Summary for 2021-06-17, created on 2021-12-20

Uniform Convergence of Interpolators: Gaussian Width, Norm Bounds, and Benign Overfitting arxiv:2106.09276 📈 244

Frederic Koehler, Lijia Zhou, Danica J. Sutherland, Nathan Srebro

**Abstract:** We consider interpolation learning in high-dimensional linear regression with Gaussian data, and prove a generic uniform convergence guarantee on the generalization error of interpolators in an arbitrary hypothesis class in terms of the class's Gaussian width. Applying the generic bound to Euclidean norm balls recovers the consistency result of Bartlett et al. (2020) for minimum-norm interpolators, and confirms a prediction of Zhou et al. (2020) for near-minimal-norm interpolators in the special case of Gaussian data. We demonstrate the generality of the bound by applying it to the simplex, obtaining a novel consistency result for minimum l1-norm interpolators (basis pursuit). Our results show how norm-based generalization bounds can explain and be used to analyze benign overfitting, at least in some settings.

XCiT: Cross-Covariance Image Transformers arxiv:2106.09681 📈 115

Alaaeldin El-Nouby, Hugo Touvron, Mathilde Caron, Piotr Bojanowski, Matthijs Douze, Armand Joulin, Ivan Laptev, Natalia Neverova, Gabriel Synnaeve, Jakob Verbeek, Hervé Jegou

**Abstract:** Following their success in natural language processing, transformers have recently shown much promise for computer vision. The self-attention operation underlying transformers yields global interactions between all tokens ,i.e. words or image patches, and enables flexible modelling of image data beyond the local interactions of convolutions. This flexibility, however, comes with a quadratic complexity in time and memory, hindering application to long sequences and high-resolution images. We propose a "transposed" version of self-attention that operates across feature channels rather than tokens, where the interactions are based on the cross-covariance matrix between keys and queries. The resulting cross-covariance attention (XCA) has linear complexity in the number of tokens, and allows efficient processing of high-resolution images. Our cross-covariance image transformer (XCiT) is built upon XCA. It combines the accuracy of conventional transformers with the scalability of convolutional architectures. We validate the effectiveness and generality of XCiT by reporting excellent results on multiple vision benchmarks, including image classification and self-supervised feature learning on ImageNet-1k, object detection and instance segmentation on COCO, and semantic segmentation on ADE20k.

Time Series is a Special Sequence: Forecasting with Sample Convolution and Interaction arxiv:2106.09305 📈 63

Minhao Liu, Ailing Zeng, Zhijian Xu, Qiuxia Lai, Qiang Xu

**Abstract:** Time series is a special type of sequence data, a set of observations collected at even time intervals and ordered chronologically. Existing deep learning techniques use generic sequence models (e.g., recurrent neural network, Transformer model, or temporal convolutional network) for time series analysis, which ignore some of its unique properties. In particular, three components characterize time series: trend, seasonality, and irregular components, and the former two components enable us to perform forecasting with reasonable accuracy. Other types of sequence data do not have such characteristics. Motivated by the above, in this paper, we propose a novel neural network architecture that conducts sample convolution and interaction for temporal modeling and apply it for the time series forecasting problem, namely \textbf{SCINet}. Compared to conventional dilated causal convolution architectures, the proposed downsample-convolve-interact architecture enables multi-resolution analysis besides expanding the receptive field of the convolution operation, which facilitates extracting temporal relation features with enhanced predictability. Experimental results show that SCINet achieves significant prediction accuracy improvement over existing solutions across various real-world time series forecasting datasets.

LNN-EL: A Neuro-Symbolic Approach to Short-text Entity Linking arxiv:2106.09795 📈 53

Hang Jiang, Sairam Gurajada, Qiuhao Lu, Sumit Neelam, Lucian Popa, Prithviraj Sen, Yunyao Li, Alexander Gray

**Abstract:** Entity linking (EL), the task of disambiguating mentions in text by linking them to entities in a knowledge graph, is crucial for text understanding, question answering or conversational systems. Entity linking on short text (e.g., single sentence or question) poses particular challenges due to limited context. While prior approaches use either heuristics or black-box neural methods, here we propose LNN-EL, a neuro-symbolic approach that combines the advantages of using interpretable rules based on first-order logic with the performance of neural learning. Even though constrained to using rules, LNN-EL performs competitively against SotA black-box neural approaches, with the added benefits of extensibility and transferability. In particular, we show that we can easily blend existing rule templates given by a human expert, with multiple types of features (priors, BERT encodings, box embeddings, etc), and even scores resulting from previous EL methods, thus improving on such methods. For instance, on the LC-QuAD-1.0 dataset, we show more than $4$\% increase in F1 score over previous SotA. Finally, we show that the inductive bias offered by using logic results in learned rules that transfer well across datasets, even without fine tuning, while maintaining high accuracy.

Learning Knowledge Graph-based World Models of Textual Environments arxiv:2106.09608 📈 51

Prithviraj Ammanabrolu, Mark O. Riedl

**Abstract:** World models improve a learning agent's ability to efficiently operate in interactive and situated environments. This work focuses on the task of building world models of text-based game environments. Text-based games, or interactive narratives, are reinforcement learning environments in which agents perceive and interact with the world using textual natural language. These environments contain long, multi-step puzzles or quests woven through a world that is filled with hundreds of characters, locations, and objects. Our world model learns to simultaneously: (1) predict changes in the world caused by an agent's actions when representing the world as a knowledge graph; and (2) generate the set of contextually relevant natural language actions required to operate in the world. We frame this task as a Set of Sequences generation problem by exploiting the inherent structure of knowledge graphs and actions and introduce both a transformer-based multi-task architecture and a loss function to train it. A zero-shot ablation study on never-before-seen textual worlds shows that our methodology significantly outperforms existing textual world modeling techniques as well as the importance of each of our contributions.

Efficient Self-supervised Vision Transformers for Representation Learning arxiv:2106.09785 📈 43

Chunyuan Li, Jianwei Yang, Pengchuan Zhang, Mei Gao, Bin Xiao, Xiyang Dai, Lu Yuan, Jianfeng Gao

**Abstract:** This paper investigates two techniques for developing efficient self-supervised vision transformers (EsViT) for visual representation learning. First, we show through a comprehensive empirical study that multi-stage architectures with sparse self-attentions can significantly reduce modeling complexity but with a cost of losing the ability to capture fine-grained correspondences between image regions. Second, we propose a new pre-training task of region matching which allows the model to capture fine-grained region dependencies and as a result significantly improves the quality of the learned vision representations. Our results show that combining the two techniques, EsViT achieves 81.3% top-1 on the ImageNet linear probe evaluation, outperforming prior arts with around an order magnitude of higher throughput. When transferring to downstream linear classification tasks, EsViT outperforms its supervised counterpart on 17 out of 18 datasets. The code and models will be publicly available.

LoRA: Low-Rank Adaptation of Large Language Models arxiv:2106.09685 📈 35

Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen

**Abstract:** An important paradigm of natural language processing consists of large-scale pre-training on general domain data and adaptation to particular tasks or domains. As we pre-train larger models, full fine-tuning, which retrains all model parameters, becomes less feasible. Using GPT-3 175B as an example -- deploying independent instances of fine-tuned models, each with 175B parameters, is prohibitively expensive. We propose Low-Rank Adaptation, or LoRA, which freezes the pre-trained model weights and injects trainable rank decomposition matrices into each layer of the Transformer architecture, greatly reducing the number of trainable parameters for downstream tasks. Compared to GPT-3 175B fine-tuned with Adam, LoRA can reduce the number of trainable parameters by 10,000 times and the GPU memory requirement by 3 times. LoRA performs on-par or better than fine-tuning in model quality on RoBERTa, DeBERTa, GPT-2, and GPT-3, despite having fewer trainable parameters, a higher training throughput, and, unlike adapters, no additional inference latency. We also provide an empirical investigation into rank-deficiency in language model adaptation, which sheds light on the efficacy of LoRA. We release a package that facilitates the integration of LoRA with PyTorch models and provide our implementations and model checkpoints for RoBERTa, DeBERTa, and GPT-2 at https://github.com/microsoft/LoRA.

Deep Learning Through the Lens of Example Difficulty arxiv:2106.09647 📈 32

Robert J. N. Baldock, Hartmut Maennel, Behnam Neyshabur

**Abstract:** Existing work on understanding deep learning often employs measures that compress all data-dependent information into a few numbers. In this work, we adopt a perspective based on the role of individual examples. We introduce a measure of the computational difficulty of making a prediction for a given input: the (effective) prediction depth. Our extensive investigation reveals surprising yet simple relationships between the prediction depth of a given input and the model's uncertainty, confidence, accuracy and speed of learning for that data point. We further categorize difficult examples into three interpretable groups, demonstrate how these groups are processed differently inside deep models and showcase how this understanding allows us to improve prediction accuracy. Insights from our study lead to a coherent view of a number of separately reported phenomena in the literature: early layers generalize while later layers memorize; early layers converge faster and networks learn easy data and simple functions first.

Multi-head or Single-head? An Empirical Comparison for Transformer Training arxiv:2106.09650 📈 25

Liyuan Liu, Jialu Liu, Jiawei Han

**Abstract:** Multi-head attention plays a crucial role in the recent success of Transformer models, which leads to consistent performance improvements over conventional attention in various applications. The popular belief is that this effectiveness stems from the ability of jointly attending multiple positions. In this paper, we first demonstrate that jointly attending multiple positions is not a unique feature of multi-head attention, as multi-layer single-head attention also attends multiple positions and is more effective. Then, we suggest the main advantage of the multi-head attention is the training stability, since it has less number of layers than the single-head attention, when attending the same number of positions. For example, 24-layer 16-head Transformer (BERT-large) and 384-layer single-head Transformer has the same total attention head number and roughly the same model size, while the multi-head one is significantly shallower. Meanwhile, we show that, with recent advances in deep learning, we can successfully stabilize the training of the 384-layer Transformer. As the training difficulty is no longer a bottleneck, substantially deeper single-head Transformer achieves consistent performance improvements without tuning hyper-parameters.

Statistical Query Lower Bounds for List-Decodable Linear Regression arxiv:2106.09689 📈 23

Ilias Diakonikolas, Daniel M. Kane, Ankit Pensia, Thanasis Pittas, Alistair Stewart

**Abstract:** We study the problem of list-decodable linear regression, where an adversary can corrupt a majority of the examples. Specifically, we are given a set $T$ of labeled examples $(x, y) \in \mathbb{R}^d \times \mathbb{R}$ and a parameter $0< α<1/2$ such that an $α$-fraction of the points in $T$ are i.i.d. samples from a linear regression model with Gaussian covariates, and the remaining $(1-α)$-fraction of the points are drawn from an arbitrary noise distribution. The goal is to output a small list of hypothesis vectors such that at least one of them is close to the target regression vector. Our main result is a Statistical Query (SQ) lower bound of $d^{\mathrm{poly}(1/α)}$ for this problem. Our SQ lower bound qualitatively matches the performance of previously developed algorithms, providing evidence that current upper bounds for this task are nearly best possible.

WaveGrad 2: Iterative Refinement for Text-to-Speech Synthesis arxiv:2106.09660 📈 21

Nanxin Chen, Yu Zhang, Heiga Zen, Ron J. Weiss, Mohammad Norouzi, Najim Dehak, William Chan

**Abstract:** This paper introduces WaveGrad 2, a non-autoregressive generative model for text-to-speech synthesis. WaveGrad 2 is trained to estimate the gradient of the log conditional density of the waveform given a phoneme sequence. The model takes an input phoneme sequence, and through an iterative refinement process, generates an audio waveform. This contrasts to the original WaveGrad vocoder which conditions on mel-spectrogram features, generated by a separate model. The iterative refinement process starts from Gaussian noise, and through a series of refinement steps (e.g., 50 steps), progressively recovers the audio sequence. WaveGrad 2 offers a natural way to trade-off between inference speed and sample quality, through adjusting the number of refinement steps. Experiments show that the model can generate high fidelity audio, approaching the performance of a state-of-the-art neural TTS system. We also report various ablation studies over different model configurations. Audio samples are available at https://wavegrad.github.io/v2.

Class Balancing GAN with a Classifier in the Loop arxiv:2106.09402 📈 20

Harsh Rangwani, Konda Reddy Mopuri, R. Venkatesh Babu

**Abstract:** Generative Adversarial Networks (GANs) have swiftly evolved to imitate increasingly complex image distributions. However, majority of the developments focus on performance of GANs on balanced datasets. We find that the existing GANs and their training regimes which work well on balanced datasets fail to be effective in case of imbalanced (i.e. long-tailed) datasets. In this work we introduce a novel theoretically motivated Class Balancing regularizer for training GANs. Our regularizer makes use of the knowledge from a pre-trained classifier to ensure balanced learning of all the classes in the dataset. This is achieved via modelling the effective class frequency based on the exponential forgetting observed in neural networks and encouraging the GAN to focus on underrepresented classes. We demonstrate the utility of our regularizer in learning representations for long-tailed distributions via achieving better performance than existing approaches over multiple datasets. Specifically, when applied to an unconditional GAN, it improves the FID from $13.03$ to $9.01$ on the long-tailed iNaturalist-$2019$ dataset.

Improving On-Screen Sound Separation for Open-Domain Videos with Audio-Visual Self-Attention arxiv:2106.09669 📈 16

Efthymios Tzinis, Scott Wisdom, Tal Remez, John R. Hershey

**Abstract:** We introduce a state-of-the-art audio-visual on-screen sound separation system which is capable of learning to separate sounds and associate them with on-screen objects by looking at in-the-wild videos. We identify limitations of previous work on audio-visual on-screen sound separation, including the simplicity and coarse resolution of spatio-temporal attention, and poor convergence of the audio separation model. Our proposed model addresses these issues using cross-modal and self-attention modules that capture audio-visual dependencies at a finer resolution over time, and by unsupervised pre-training of audio separation model. These improvements allow the model to generalize to a much wider set of unseen videos. We also show a robust way to further improve the generalization capability of our models by calibrating the probabilities of our audio-visual on-screen classifier, using only a small amount of in-domain videos labeled for their on-screen presence. For evaluation and semi-supervised training, we collected human annotations of on-screen audio from a large database of in-the-wild videos (YFCC100m). Our results show marked improvements in on-screen separation performance, in more general conditions than previous methods.

Biomedical Interpretable Entity Representations arxiv:2106.09502 📈 15

Diego Garcia-Olano, Yasumasa Onoe, Ioana Baldini, Joydeep Ghosh, Byron C. Wallace, Kush R. Varshney

**Abstract:** Pre-trained language models induce dense entity representations that offer strong performance on entity-centric NLP tasks, but such representations are not immediately interpretable. This can be a barrier to model uptake in important domains such as biomedicine. There has been recent work on general interpretable representation learning (Onoe and Durrett, 2020), but these domain-agnostic representations do not readily transfer to the important domain of biomedicine. In this paper, we create a new entity type system and training set from a large corpus of biomedical texts by mapping entities to concepts in a medical ontology, and from these to Wikipedia pages whose categories are our types. From this mapping we derive Biomedical Interpretable Entity Representations(BIERs), in which dimensions correspond to fine-grained entity types, and values are predicted probabilities that a given entity is of the corresponding type. We propose a novel method that exploits BIER's final sparse and intermediate dense representations to facilitate model and entity type debugging. We show that BIERs achieve strong performance in biomedical tasks including named entity disambiguation and entity label classification, and we provide error analysis to highlight the utility of their interpretability, particularly in low-supervision settings. Finally, we provide our induced 68K biomedical type system, the corresponding 37 million triples of derived data used to train BIER models and our best performing model.

Multi-Label Learning from Single Positive Labels arxiv:2106.09708 📈 14

Elijah Cole, Oisin Mac Aodha, Titouan Lorieul, Pietro Perona, Dan Morris, Nebojsa Jojic

**Abstract:** Predicting all applicable labels for a given image is known as multi-label classification. Compared to the standard multi-class case (where each image has only one label), it is considerably more challenging to annotate training data for multi-label classification. When the number of potential labels is large, human annotators find it difficult to mention all applicable labels for each training image. Furthermore, in some settings detection is intrinsically difficult e.g. finding small object instances in high resolution images. As a result, multi-label training data is often plagued by false negatives. We consider the hardest version of this problem, where annotators provide only one relevant label for each image. As a result, training sets will have only one positive label per image and no confirmed negatives. We explore this special case of learning from missing labels across four different multi-label image classification datasets for both linear classifiers and end-to-end fine-tuned deep networks. We extend existing multi-label losses to this setting and propose novel variants that constrain the number of expected positive labels during training. Surprisingly, we show that in some cases it is possible to approach the performance of fully labeled classifiers despite training with significantly fewer confirmed labels.

Always Be Dreaming: A New Approach for Data-Free Class-Incremental Learning arxiv:2106.09701 📈 13

James Smith, Yen-Chang Hsu, Jonathan Balloch, Yilin Shen, Hongxia Jin, Zsolt Kira

**Abstract:** Modern computer vision applications suffer from catastrophic forgetting when incrementally learning new concepts over time. The most successful approaches to alleviate this forgetting require extensive replay of previously seen data, which is problematic when memory constraints or data legality concerns exist. In this work, we consider the high-impact problem of Data-Free Class-Incremental Learning (DFCIL), where an incremental learning agent must learn new concepts over time without storing generators or training data from past tasks. One approach for DFCIL is to replay synthetic images produced by inverting a frozen copy of the learner's classification model, but we show this approach fails for common class-incremental benchmarks when using standard distillation strategies. We diagnose the cause of this failure and propose a novel incremental distillation strategy for DFCIL, contributing a modified cross-entropy training and importance-weighted feature distillation, and show that our method results in up to a 25.1% increase in final task accuracy (absolute difference) compared to SOTA DFCIL methods for common class-incremental benchmarks. Our method even outperforms several standard replay based methods which store a coreset of images.

How Low Can We Go: Trading Memory for Error in Low-Precision Training arxiv:2106.09686 📈 13

Chengrun Yang, Ziyang Wu, Jerry Chee, Christopher De Sa, Madeleine Udell

**Abstract:** Low-precision arithmetic trains deep learning models using less energy, less memory and less time. However, we pay a price for the savings: lower precision may yield larger round-off error and hence larger prediction error. As applications proliferate, users must choose which precision to use to train a new model, and chip manufacturers must decide which precisions to manufacture. We view these precision choices as a hyperparameter tuning problem, and borrow ideas from meta-learning to learn the tradeoff between memory and error. In this paper, we introduce Pareto Estimation to Pick the Perfect Precision (PEPPP). We use matrix factorization to find non-dominated configurations (the Pareto frontier) with a limited number of network evaluations. For any given memory budget, the precision that minimizes error is a point on this frontier. Practitioners can use the frontier to trade memory for error and choose the best precision for their goals.

Scientific Language Models for Biomedical Knowledge Base Completion: An Empirical Study arxiv:2106.09700 📈 12

Rahul Nadkarni, David Wadden, Iz Beltagy, Noah A. Smith, Hannaneh Hajishirzi, Tom Hope

**Abstract:** Biomedical knowledge graphs (KGs) hold rich information on entities such as diseases, drugs, and genes. Predicting missing links in these graphs can boost many important applications, such as drug design and repurposing. Recent work has shown that general-domain language models (LMs) can serve as "soft" KGs, and that they can be fine-tuned for the task of KG completion. In this work, we study scientific LMs for KG completion, exploring whether we can tap into their latent knowledge to enhance biomedical link prediction. We evaluate several domain-specific LMs, fine-tuning them on datasets centered on drugs and diseases that we represent as KGs and enrich with textual entity descriptions. We integrate the LM-based models with KG embedding models, using a router method that learns to assign each input example to either type of model and provides a substantial boost in performance. Finally, we demonstrate the advantage of LM models in the inductive setting with novel scientific entities. Our datasets and code are made publicly available.

Adapting the Function Approximation Architecture in Online Reinforcement Learning arxiv:2106.09776 📈 11

John D. Martin, Joseph Modayil

**Abstract:** The performance of a reinforcement learning (RL) system depends on the computational architecture used to approximate a value function. Deep learning methods provide both optimization techniques and architectures for approximating nonlinear functions from noisy, high-dimensional observations. However, prevailing optimization techniques are not designed for strictly-incremental online updates. Nor are standard architectures designed for observations with an a priori unknown structure: for example, light sensors randomly dispersed in space. This paper proposes an online RL prediction algorithm with an adaptive architecture that efficiently finds useful nonlinear features. The algorithm is evaluated in a spatial domain with high-dimensional, stochastic observations. The algorithm outperforms non-adaptive baseline architectures and approaches the performance of an architecture given side-channel information. These results are a step towards scalable RL algorithms for more general problems, where the observation structure is not available.

SECANT: Self-Expert Cloning for Zero-Shot Generalization of Visual Policies arxiv:2106.09678 📈 11

Linxi Fan, Guanzhi Wang, De-An Huang, Zhiding Yu, Li Fei-Fei, Yuke Zhu, Anima Anandkumar

**Abstract:** Generalization has been a long-standing challenge for reinforcement learning (RL). Visual RL, in particular, can be easily distracted by irrelevant factors in high-dimensional observation space. In this work, we consider robust policy learning which targets zero-shot generalization to unseen visual environments with large distributional shift. We propose SECANT, a novel self-expert cloning technique that leverages image augmentation in two stages to decouple robust representation learning from policy optimization. Specifically, an expert policy is first trained by RL from scratch with weak augmentations. A student network then learns to mimic the expert policy by supervised learning with strong augmentations, making its representation more robust against visual variations compared to the expert. Extensive experiments demonstrate that SECANT significantly advances the state of the art in zero-shot generalization across 4 challenging domains. Our average reward improvements over prior SOTAs are: DeepMind Control (+26.5%), robotic manipulation (+337.8%), vision-based autonomous driving (+47.7%), and indoor object navigation (+15.8%). Code release and video are available at https://linxifan.github.io/secant-site/.

PyKale: Knowledge-Aware Machine Learning from Multiple Sources in Python arxiv:2106.09756 📈 10

Haiping Lu, Xianyuan Liu, Robert Turner, Peizhen Bai, Raivo E Koot, Shuo Zhou, Mustafa Chasmai, Lawrence Schobs

**Abstract:** Machine learning is a general-purpose technology holding promises for many interdisciplinary research problems. However, significant barriers exist in crossing disciplinary boundaries when most machine learning tools are developed in different areas separately. We present Pykale - a Python library for knowledge-aware machine learning on graphs, images, texts, and videos to enable and accelerate interdisciplinary research. We formulate new green machine learning guidelines based on standard software engineering practices and propose a novel pipeline-based application programming interface (API). PyKale focuses on leveraging knowledge from multiple sources for accurate and interpretable prediction, thus supporting multimodal learning and transfer learning (particularly domain adaptation) with latest deep learning and dimensionality reduction models. We build PyKale on PyTorch and leverage the rich PyTorch ecosystem. Our pipeline-based API design enforces standardization and minimalism, embracing green machine learning concepts via reducing repetitions and redundancy, reusing existing resources, and recycling learning models across areas. We demonstrate its interdisciplinary nature via examples in bioinformatics, knowledge graph, image/video recognition, and medical imaging.

Voice2Series: Reprogramming Acoustic Models for Time Series Classification arxiv:2106.09296 📈 10

Chao-Han Huck Yang, Yun-Yun Tsai, Pin-Yu Chen

**Abstract:** Learning to classify time series with limited data is a practical yet challenging problem. Current methods are primarily based on hand-designed feature extraction rules or domain-specific data augmentation. Motivated by the advances in deep speech processing models and the fact that voice data are univariate temporal signals, in this paper, we propose Voice2Series (V2S), a novel end-to-end approach that reprograms acoustic models for time series classification, through input transformation learning and output label mapping. Leveraging the representation learning power of a large-scale pre-trained speech processing model, on 30 different time series tasks we show that V2S either outperforms or is tied with state-of-the-art methods on 20 tasks, and improves their average accuracy by 1.84%. We further provide a theoretical justification of V2S by proving its population risk is upper bounded by the source risk and a Wasserstein distance accounting for feature alignment via reprogramming. Our results offer new and effective means to time series classification.

Iterative Feature Matching: Toward Provable Domain Generalization with Logarithmic Environments arxiv:2106.09913 📈 9

Yining Chen, Elan Rosenfeld, Mark Sellke, Tengyu Ma, Andrej Risteski

**Abstract:** Domain generalization aims at performing well on unseen test environments with data from a limited number of training environments. Despite a proliferation of proposal algorithms for this task, assessing their performance both theoretically and empirically is still very challenging. Distributional matching algorithms such as (Conditional) Domain Adversarial Networks [Ganin et al., 2016, Long et al., 2018] are popular and enjoy empirical success, but they lack formal guarantees. Other approaches such as Invariant Risk Minimization (IRM) require a prohibitively large number of training environments -- linear in the dimension of the spurious feature space $d_s$ -- even on simple data models like the one proposed by [Rosenfeld et al., 2021]. Under a variant of this model, we show that both ERM and IRM cannot generalize with $o(d_s)$ environments. We then present an iterative feature matching algorithm that is guaranteed with high probability to yield a predictor that generalizes after seeing only $O(\log d_s)$ environments. Our results provide the first theoretical justification for a family of distribution-matching algorithms widely used in practice under a concrete nontrivial data model.

BABEL: Bodies, Action and Behavior with English Labels arxiv:2106.09696 📈 9

Abhinanda R. Punnakkal, Arjun Chandrasekaran, Nikos Athanasiou, Alejandra Quiros-Ramirez, Michael J. Black

**Abstract:** Understanding the semantics of human movement -- the what, how and why of the movement -- is an important problem that requires datasets of human actions with semantic labels. Existing datasets take one of two approaches. Large-scale video datasets contain many action labels but do not contain ground-truth 3D human motion. Alternatively, motion-capture (mocap) datasets have precise body motions but are limited to a small number of actions. To address this, we present BABEL, a large dataset with language labels describing the actions being performed in mocap sequences. BABEL consists of action labels for about 43 hours of mocap sequences from AMASS. Action labels are at two levels of abstraction -- sequence labels describe the overall action in the sequence, and frame labels describe all actions in every frame of the sequence. Each frame label is precisely aligned with the duration of the corresponding action in the mocap sequence, and multiple actions can overlap. There are over 28k sequence labels, and 63k frame labels in BABEL, which belong to over 250 unique action categories. Labels from BABEL can be leveraged for tasks like action recognition, temporal action localization, motion synthesis, etc. To demonstrate the value of BABEL as a benchmark, we evaluate the performance of models on 3D action recognition. We demonstrate that BABEL poses interesting learning challenges that are applicable to real-world scenarios, and can serve as a useful benchmark of progress in 3D action recognition. The dataset, baseline method, and evaluation code is made available, and supported for academic research purposes at https://babel.is.tue.mpg.de/.

Modeling Worlds in Text arxiv:2106.09578 📈 9

Prithviraj Ammanabrolu, Mark O. Riedl

**Abstract:** We provide a dataset that enables the creation of learning agents that can build knowledge graph-based world models of interactive narratives. Interactive narratives -- or text-adventure games -- are partially observable environments structured as long puzzles or quests in which an agent perceives and interacts with the world purely through textual natural language. Each individual game typically contains hundreds of locations, characters, and objects -- each with their own unique descriptions -- providing an opportunity to study the problem of giving language-based agents the structured memory necessary to operate in such worlds. Our dataset provides 24198 mappings between rich natural language observations and: (1) knowledge graphs that reflect the world state in the form of a map; (2) natural language actions that are guaranteed to cause a change in that particular world state. The training data is collected across 27 games in multiple genres and contains a further 7836 heldout instances over 9 additional games in the test set. We further provide baseline models using rules-based, question-answering, and sequence learning approaches in addition to an analysis of the data and corresponding learning tasks.

Pruning Randomly Initialized Neural Networks with Iterative Randomization arxiv:2106.09269 📈 9

Daiki Chijiwa, Shin'ya Yamaguchi, Yasutoshi Ida, Kenji Umakoshi, Tomohiro Inoue

**Abstract:** Pruning the weights of randomly initialized neural networks plays an important role in the context of lottery ticket hypothesis. Ramanujan et al. (2020) empirically showed that only pruning the weights can achieve remarkable performance instead of optimizing the weight values. However, to achieve the same level of performance as the weight optimization, the pruning approach requires more parameters in the networks before pruning and thus more memory space. To overcome this parameter inefficiency, we introduce a novel framework to prune randomly initialized neural networks with iteratively randomizing weight values (IteRand). Theoretically, we prove an approximation theorem in our framework, which indicates that the randomizing operations are provably effective to reduce the required number of the parameters. We also empirically demonstrate the parameter efficiency in multiple experiments on CIFAR-10 and ImageNet.

AI-Enabled Ultra-Low-Dose CT Reconstruction arxiv:2106.09834 📈 8

Weiwen Wu, Chuang Niu, Shadi Ebrahimian, Hengyong Yu, Mannu Kalra, Ge Wang

**Abstract:** By the ALARA (As Low As Reasonably Achievable) principle, ultra-low-dose CT reconstruction is a holy grail to minimize cancer risks and genetic damages, especially for children. With the development of medical CT technologies, the iterative algorithms are widely used to reconstruct decent CT images from a low-dose scan. Recently, artificial intelligence (AI) techniques have shown a great promise in further reducing CT radiation dose to the next level. In this paper, we demonstrate that AI-powered CT reconstruction offers diagnostic image quality at an ultra-low-dose level comparable to that of radiography. Specifically, here we develop a Split Unrolled Grid-like Alternative Reconstruction (SUGAR) network, in which deep learning, physical modeling and image prior are integrated. The reconstruction results from clinical datasets show that excellent images can be reconstructed using SUGAR from 36 projections. This approach has a potential to change future healthcare.

Hybrid graph convolutional neural networks for landmark-based anatomical segmentation arxiv:2106.09832 📈 8

Nicolás Gaggion, Lucas Mansilla, Diego Milone, Enzo Ferrante

**Abstract:** In this work we address the problem of landmark-based segmentation for anatomical structures. We propose HybridGNet, an encoder-decoder neural architecture which combines standard convolutions for image feature encoding, with graph convolutional neural networks to decode plausible representations of anatomical structures. We benchmark the proposed architecture considering other standard landmark and pixel-based models for anatomical segmentation in chest x-ray images, and found that HybridGNet is more robust to image occlusions. We also show that it can be used to construct landmark-based segmentations from pixel level annotations. Our experimental results suggest that HybridGNet produces accurate and anatomically plausible landmark-based segmentations, by naturally incorporating shape constraints within the decoding process via spectral convolutions.

Unsupervised Resource Allocation with Graph Neural Networks arxiv:2106.09761 📈 8

Miles Cranmer, Peter Melchior, Brian Nord

**Abstract:** We present an approach for maximizing a global utility function by learning how to allocate resources in an unsupervised way. We expect interactions between allocation targets to be important and therefore propose to learn the reward structure for near-optimal allocation policies with a GNN. By relaxing the resource constraint, we can employ gradient-based optimization in contrast to more standard evolutionary algorithms. Our algorithm is motivated by a problem in modern astronomy, where one needs to select-based on limited initial information-among $10^9$ galaxies those whose detailed measurement will lead to optimal inference of the composition of the universe. Our technique presents a way of flexibly learning an allocation strategy by only requiring forward simulators for the physics of interest and the measurement process. We anticipate that our technique will also find applications in a range of resource allocation problems.

Orthogonal-Padé Activation Functions: Trainable Activation functions for smooth and faster convergence in deep networks arxiv:2106.09693 📈 8

Koushik Biswas, Shilpak Banerjee, Ashish Kumar Pandey

**Abstract:** We have proposed orthogonal-Padé activation functions, which are trainable activation functions and show that they have faster learning capability and improves the accuracy in standard deep learning datasets and models. Based on our experiments, we have found two best candidates out of six orthogonal-Padé activations, which we call safe Hermite-Pade (HP) activation functions, namely HP-1 and HP-2. When compared to ReLU, HP-1 and HP-2 has an increment in top-1 accuracy by 5.06% and 4.63% respectively in PreActResNet-34, by 3.02% and 2.75% respectively in MobileNet V2 model on CIFAR100 dataset while on CIFAR10 dataset top-1 accuracy increases by 2.02% and 1.78% respectively in PreActResNet-34, by 2.24% and 2.06% respectively in LeNet, by 2.15% and 2.03% respectively in Efficientnet B0.

PAC-Bayes, MAC-Bayes and Conditional Mutual Information: Fast rate bounds that handle general VC classes arxiv:2106.09683 📈 8

Peter Grünwald, Thomas Steinke, Lydia Zakynthinou

**Abstract:** We give a novel, unified derivation of conditional PAC-Bayesian and mutual information (MI) generalization bounds. We derive conditional MI bounds as an instance, with special choice of prior, of conditional MAC-Bayesian (Mean Approximately Correct) bounds, itself derived from conditional PAC-Bayesian bounds, where `conditional' means that one can use priors conditioned on a joint training and ghost sample. This allows us to get nontrivial PAC-Bayes and MI-style bounds for general VC classes, something recently shown to be impossible with standard PAC-Bayesian/MI bounds. Second, it allows us to get faster rates of order $O \left(({\text{KL}}/n)^γ\right)$ for $γ> 1/2$ if a Bernstein condition holds and for exp-concave losses (with $γ=1$), which is impossible with both standard PAC-Bayes generalization and MI bounds. Our work extends the recent work by Steinke and Zakynthinou [2020] who handle MI with VC but neither PAC-Bayes nor fast rates, the recent work of Hellström and Durisi [2020] who extend the latter to the PAC-Bayes setting via a unifying exponential inequality, and Mhammedi et al. [2019] who initiated fast rate PAC-Bayes generalization error bounds but handle neither MI nor general VC classes.

Algorithmic Bias and Data Bias: Understanding the Relation between Distributionally Robust Optimization and Data Curation arxiv:2106.09467 📈 8

Agnieszka Słowik, Léon Bottou

**Abstract:** Machine learning systems based on minimizing average error have been shown to perform inconsistently across notable subsets of the data, which is not exposed by a low average error for the entire dataset. In consequential social and economic applications, where data represent people, this can lead to discrimination of underrepresented gender and ethnic groups. Given the importance of bias mitigation in machine learning, the topic leads to contentious debates on how to ensure fairness in practice (data bias versus algorithmic bias). Distributionally Robust Optimization (DRO) seemingly addresses this problem by minimizing the worst expected risk across subpopulations. We establish theoretical results that clarify the relation between DRO and the optimization of the same loss averaged on an adequately weighted training dataset. The results cover finite and infinite number of training distributions, as well as convex and non-convex loss functions. We show that neither DRO nor curating the training set should be construed as a complete solution for bias mitigation: in the same way that there is no universally robust training set, there is no universal way to setup a DRO problem and ensure a socially acceptable set of results. We then leverage these insights to provide a mininal set of practical recommendations for addressing bias with DRO. Finally, we discuss ramifications of our results in other related applications of DRO, using an example of adversarial robustness. Our results show that there is merit to both the algorithm-focused and the data-focused side of the bias debate, as long as arguments in favor of these positions are precisely qualified and backed by relevant mathematics known today.

Deep Subdomain Adaptation Network for Image Classification arxiv:2106.09388 📈 8

Yongchun Zhu, Fuzhen Zhuang, Jindong Wang, Guolin Ke, Jingwu Chen, Jiang Bian, Hui Xiong, Qing He

**Abstract:** For a target task where labeled data is unavailable, domain adaptation can transfer a learner from a different source domain. Previous deep domain adaptation methods mainly learn a global domain shift, i.e., align the global source and target distributions without considering the relationships between two subdomains within the same category of different domains, leading to unsatisfying transfer learning performance without capturing the fine-grained information. Recently, more and more researchers pay attention to Subdomain Adaptation which focuses on accurately aligning the distributions of the relevant subdomains. However, most of them are adversarial methods which contain several loss functions and converge slowly. Based on this, we present Deep Subdomain Adaptation Network (DSAN) which learns a transfer network by aligning the relevant subdomain distributions of domain-specific layer activations across different domains based on a local maximum mean discrepancy (LMMD). Our DSAN is very simple but effective which does not need adversarial training and converges fast. The adaptation can be achieved easily with most feed-forward network models by extending them with LMMD loss, which can be trained efficiently via back-propagation. Experiments demonstrate that DSAN can achieve remarkable results on both object recognition tasks and digit classification tasks. Our code will be available at: https://github.com/easezyc/deep-transfer-learning

Invisible for both Camera and LiDAR: Security of Multi-Sensor Fusion based Perception in Autonomous Driving Under Physical-World Attacks arxiv:2106.09249 📈 8

Yulong Cao*, Ningfei Wang*, Chaowei Xiao*, Dawei Yang*, Jin Fang, Ruigang Yang, Qi Alfred Chen, Mingyan Liu, Bo Li

**Abstract:** In Autonomous Driving (AD) systems, perception is both security and safety critical. Despite various prior studies on its security issues, all of them only consider attacks on camera- or LiDAR-based AD perception alone. However, production AD systems today predominantly adopt a Multi-Sensor Fusion (MSF) based design, which in principle can be more robust against these attacks under the assumption that not all fusion sources are (or can be) attacked at the same time. In this paper, we present the first study of security issues of MSF-based perception in AD systems. We directly challenge the basic MSF design assumption above by exploring the possibility of attacking all fusion sources simultaneously. This allows us for the first time to understand how much security guarantee MSF can fundamentally provide as a general defense strategy for AD perception. We formulate the attack as an optimization problem to generate a physically-realizable, adversarial 3D-printed object that misleads an AD system to fail in detecting it and thus crash into it. We propose a novel attack pipeline that addresses two main design challenges: (1) non-differentiable target camera and LiDAR sensing systems, and (2) non-differentiable cell-level aggregated features popularly used in LiDAR-based AD perception. We evaluate our attack on MSF included in representative open-source industry-grade AD systems in real-world driving scenarios. Our results show that the attack achieves over 90% success rate across different object types and MSF. Our attack is also found stealthy, robust to victim positions, transferable across MSF algorithms, and physical-world realizable after being 3D-printed and captured by LiDAR and camera devices. To concretely assess the end-to-end safety impact, we further perform simulation evaluation and show that it can cause a 100% vehicle collision rate for an industry-grade AD system.

Bad Characters: Imperceptible NLP Attacks arxiv:2106.09898 📈 7

Nicholas Boucher, Ilia Shumailov, Ross Anderson, Nicolas Papernot

**Abstract:** Several years of research have shown that machine-learning systems are vulnerable to adversarial examples, both in theory and in practice. Until now, such attacks have primarily targeted visual models, exploiting the gap between human and machine perception. Although text-based models have also been attacked with adversarial examples, such attacks struggled to preserve semantic meaning and indistinguishability. In this paper, we explore a large class of adversarial examples that can be used to attack text-based models in a black-box setting without making any human-perceptible visual modification to inputs. We use encoding-specific perturbations that are imperceptible to the human eye to manipulate the outputs of a wide range of Natural Language Processing (NLP) systems from neural machine-translation pipelines to web search engines. We find that with a single imperceptible encoding injection -- representing one invisible character, homoglyph, reordering, or deletion -- an attacker can significantly reduce the performance of vulnerable models, and with three injections most models can be functionally broken. Our attacks work against currently-deployed commercial systems, including those produced by Microsoft and Google, in addition to open source models published by Facebook, IBM, and HuggingFace. This novel series of attacks presents a significant threat to many language processing systems: an attacker can affect systems in a targeted manner without any assumptions about the underlying model. We conclude that text-based NLP systems require careful input sanitization, just like conventional applications, and that given such systems are now being deployed rapidly at scale, the urgent attention of architects and operators is required.

Dual-Teacher Class-Incremental Learning With Data-Free Generative Replay arxiv:2106.09835 📈 7

Yoojin Choi, Mostafa El-Khamy, Jungwon Lee

**Abstract:** This paper proposes two novel knowledge transfer techniques for class-incremental learning (CIL). First, we propose data-free generative replay (DF-GR) to mitigate catastrophic forgetting in CIL by using synthetic samples from a generative model. In the conventional generative replay, the generative model is pre-trained for old data and shared in extra memory for later incremental learning. In our proposed DF-GR, we train a generative model from scratch without using any training data, based on the pre-trained classification model from the past, so we curtail the cost of sharing pre-trained generative models. Second, we introduce dual-teacher information distillation (DT-ID) for knowledge distillation from two teachers to one student. In CIL, we use DT-ID to learn new classes incrementally based on the pre-trained model for old classes and another model (pre-)trained on the new data for new classes. We implemented the proposed schemes on top of one of the state-of-the-art CIL methods and showed the performance improvement on CIFAR-100 and ImageNet datasets.

Causal Bias Quantification for Continuous Treatment arxiv:2106.09762 📈 7

Gianluca Detommaso, Michael Brückner, Philip Schulz, Victor Chernozhukov

**Abstract:** In this work we develop a novel characterization of marginal causal effect and causal bias in the continuous treatment setting. We show they can be expressed as an expectation with respect to a conditional probability distribution, which can be estimated via standard statistical and probabilistic methods. All terms in the expectations can be computed via automatic differentiation, also for highly non-linear models. We further develop a new complete criterion for identifiability of causal effects via covariate adjustment, showing the bias equals zero if the criterion is met. We study the effectiveness of our framework in three different scenarios: linear models under confounding, overcontrol and endogenous selection bias; a non-linear model where full identifiability cannot be achieved because of missing data; a simulated medical study of statins and atherosclerotic cardiovascular disease.

Adaptive Low-Rank Regularization with Damping Sequences to Restrict Lazy Weights in Deep Networks arxiv:2106.09677 📈 7

Mohammad Mahdi Bejani, Mehdi Ghatee

**Abstract:** Overfitting is one of the critical problems in deep neural networks. Many regularization schemes try to prevent overfitting blindly. However, they decrease the convergence speed of training algorithms. Adaptive regularization schemes can solve overfitting more intelligently. They usually do not affect the entire network weights. This paper detects a subset of the weighting layers that cause overfitting. The overfitting recognizes by matrix and tensor condition numbers. An adaptive regularization scheme entitled Adaptive Low-Rank (ALR) is proposed that converges a subset of the weighting layers to their Low-Rank Factorization (LRF). It happens by minimizing a new Tikhonov-based loss function. ALR also encourages lazy weights to contribute to the regularization when epochs grow up. It uses a damping sequence to increment layer selection likelihood in the last generations. Thus before falling the training accuracy, ALR reduces the lazy weights and regularizes the network substantially. The experimental results show that ALR regularizes the deep networks well with high training speed and low resource usage.

CRIL: Continual Robot Imitation Learning via Generative and Prediction Model arxiv:2106.09422 📈 7

Chongkai Gao, Haichuan Gao, Shangqi Guo, Tianren Zhang, Feng Chen

**Abstract:** Imitation learning (IL) algorithms have shown promising results for robots to learn skills from expert demonstrations. However, they need multi-task demonstrations to be provided at once for acquiring diverse skills, which is difficult in real world. In this work we study how to realize continual imitation learning ability that empowers robots to continually learn new tasks one by one, thus reducing the burden of multi-task IL and accelerating the process of new task learning at the same time. We propose a novel trajectory generation model that employs both a generative adversarial network and a dynamics-aware prediction model to generate pseudo trajectories from all learned tasks in the new task learning process. Our experiments on both simulation and real-world manipulation tasks demonstrate the effectiveness of our method.

A Unified Generative Adversarial Network Training via Self-Labeling and Self-Attention arxiv:2106.09914 📈 6

Tomoki Watanabe, Paolo Favaro

**Abstract:** We propose a novel GAN training scheme that can handle any level of labeling in a unified manner. Our scheme introduces a form of artificial labeling that can incorporate manually defined labels, when available, and induce an alignment between them. To define the artificial labels, we exploit the assumption that neural network generators can be trained more easily to map nearby latent vectors to data with semantic similarities, than across separate categories. We use generated data samples and their corresponding artificial conditioning labels to train a classifier. The classifier is then used to self-label real data. To boost the accuracy of the self-labeling, we also use the exponential moving average of the classifier. However, because the classifier might still make mistakes, especially at the beginning of the training, we also refine the labels through self-attention, by using the labeling of real data samples only when the classifier outputs a high classification probability score. We evaluate our approach on CIFAR-10, STL-10 and SVHN, and show that both self-labeling and self-attention consistently improve the quality of generated data. More surprisingly, we find that the proposed scheme can even outperform class-conditional GANs.

Multi-Task Learning and Adapted Knowledge Models for Emotion-Cause Extraction arxiv:2106.09790 📈 6

Elsbeth Turcan, Shuai Wang, Rishita Anubhai, Kasturi Bhattacharjee, Yaser Al-Onaizan, Smaranda Muresan

**Abstract:** Detecting what emotions are expressed in text is a well-studied problem in natural language processing. However, research on finer grained emotion analysis such as what causes an emotion is still in its infancy. We present solutions that tackle both emotion recognition and emotion cause detection in a joint fashion. Considering that common-sense knowledge plays an important role in understanding implicitly expressed emotions and the reasons for those emotions, we propose novel methods that combine common-sense knowledge via adapted knowledge models with multi-task learning to perform joint emotion classification and emotion cause tagging. We show performance improvement on both tasks when including common-sense reasoning and a multitask framework. We provide a thorough analysis to gain insights into model performance.

Guided Integrated Gradients: An Adaptive Path Method for Removing Noise arxiv:2106.09788 📈 6

Andrei Kapishnikov, Subhashini Venugopalan, Besim Avci, Ben Wedin, Michael Terry, Tolga Bolukbasi

**Abstract:** Integrated Gradients (IG) is a commonly used feature attribution method for deep neural networks. While IG has many desirable properties, the method often produces spurious/noisy pixel attributions in regions that are not related to the predicted class when applied to visual models. While this has been previously noted, most existing solutions are aimed at addressing the symptoms by explicitly reducing the noise in the resulting attributions. In this work, we show that one of the causes of the problem is the accumulation of noise along the IG path. To minimize the effect of this source of noise, we propose adapting the attribution path itself -- conditioning the path not just on the image but also on the model being explained. We introduce Adaptive Path Methods (APMs) as a generalization of path methods, and Guided IG as a specific instance of an APM. Empirically, Guided IG creates saliency maps better aligned with the model's prediction and the input image that is being explained. We show through qualitative and quantitative experiments that Guided IG outperforms other, related methods in nearly every experiment.

Spectral goodness-of-fit tests for complete and partial network data arxiv:2106.09702 📈 6

Shane Lubold, Bolun Liu, Tyler H. McCormick

**Abstract:** Networks describe the, often complex, relationships between individual actors. In this work, we address the question of how to determine whether a parametric model, such as a stochastic block model or latent space model, fits a dataset well and will extrapolate to similar data. We use recent results in random matrix theory to derive a general goodness-of-fit test for dyadic data. We show that our method, when applied to a specific model of interest, provides an straightforward, computationally fast way of selecting parameters in a number of commonly used network models. For example, we show how to select the dimension of the latent space in latent space models. Unlike other network goodness-of-fit methods, our general approach does not require simulating from a candidate parametric model, which can be cumbersome with large graphs, and eliminates the need to choose a particular set of statistics on the graph for comparison. It also allows us to perform goodness-of-fit tests on partial network data, such as Aggregated Relational Data. We show with simulations that our method performs well in many situations of interest. We analyze several empirically relevant networks and show that our method leads to improved community detection algorithms. R code to implement our method is available on Github.

Knowledge distillation from multi-modal to mono-modal segmentation networks arxiv:2106.09564 📈 6

Minhao Hu, Matthis Maillard, Ya Zhang, Tommaso Ciceri, Giammarco La Barbera, Isabelle Bloch, Pietro Gori

**Abstract:** The joint use of multiple imaging modalities for medical image segmentation has been widely studied in recent years. The fusion of information from different modalities has demonstrated to improve the segmentation accuracy, with respect to mono-modal segmentations, in several applications. However, acquiring multiple modalities is usually not possible in a clinical setting due to a limited number of physicians and scanners, and to limit costs and scan time. Most of the time, only one modality is acquired. In this paper, we propose KD-Net, a framework to transfer knowledge from a trained multi-modal network (teacher) to a mono-modal one (student). The proposed method is an adaptation of the generalized distillation framework where the student network is trained on a subset (1 modality) of the teacher's inputs (n modalities). We illustrate the effectiveness of the proposed framework in brain tumor segmentation with the BraTS 2018 dataset. Using different architectures, we show that the student network effectively learns from the teacher and always outperforms the baseline mono-modal network in terms of segmentation accuracy.

Optimising simulations for diphoton production at hadron colliders using amplitude neural networks arxiv:2106.09474 📈 6

Joseph Aylett-Bullock, Simon Badger, Ryan Moodie

**Abstract:** Machine learning technology has the potential to dramatically optimise event generation and simulations. We continue to investigate the use of neural networks to approximate matrix elements for high-multiplicity scattering processes. We focus on the case of loop-induced diphoton production through gluon fusion and develop a realistic simulation method that can be applied to hadron collider observables. Neural networks are trained using the one-loop amplitudes implemented in the NJet C++ library and interfaced to the Sherpa Monte Carlo event generator where we perform a detailed study for $2\to3$ and $2\to4$ scattering problems. We also consider how the trained networks perform when varying the kinematic cuts effecting the phase space and the reliability of the neural network simulations.

Multi-Agent Training beyond Zero-Sum with Correlated Equilibrium Meta-Solvers arxiv:2106.09435 📈 6

Luke Marris, Paul Muller, Marc Lanctot, Karl Tuyls, Thore Graepel

**Abstract:** Two-player, constant-sum games are well studied in the literature, but there has been limited progress outside of this setting. We propose Joint Policy-Space Response Oracles (JPSRO), an algorithm for training agents in n-player, general-sum extensive form games, which provably converges to an equilibrium. We further suggest correlated equilibria (CE) as promising meta-solvers, and propose a novel solution concept Maximum Gini Correlated Equilibrium (MGCE), a principled and computationally efficient family of solutions for solving the correlated equilibrium selection problem. We conduct several experiments using CE meta-solvers for JPSRO and demonstrate convergence on n-player, general-sum games.

Trainable Discrete Feature Embeddings for Variational Quantum Classifier arxiv:2106.09415 📈 6

Napat Thumwanit, Chayaphol Lortaraprasert, Hiroshi Yano, Rudy Raymond

**Abstract:** Quantum classifiers provide sophisticated embeddings of input data in Hilbert space promising quantum advantage. The advantage stems from quantum feature maps encoding the inputs into quantum states with variational quantum circuits. A recent work shows how to map discrete features with fewer quantum bits using Quantum Random Access Coding (QRAC), an important primitive to encode binary strings into quantum states. We propose a new method to embed discrete features with trainable quantum circuits by combining QRAC and a recently proposed strategy for training quantum feature map called quantum metric learning. We show that the proposed trainable embedding requires not only as few qubits as QRAC but also overcomes the limitations of QRAC to classify inputs whose classes are based on hard Boolean functions. We numerically demonstrate its use in variational quantum classifiers to achieve better performances in classifying real-world datasets, and thus its possibility to leverage near-term quantum computers for quantum machine learning.

Quantized Federated Learning under Transmission Delay and Outage Constraints arxiv:2106.09397 📈 6

Yanmeng Wang, Yanqing Xu, Qingjiang Shi, Tsung-Hui Chang

**Abstract:** Federated learning (FL) has been recognized as a viable distributed learning paradigm which trains a machine learning model collaboratively with massive mobile devices in the wireless edge while protecting user privacy. Although various communication schemes have been proposed to expedite the FL process, most of them have assumed ideal wireless channels which provide reliable and lossless communication links between the server and mobile clients. Unfortunately, in practical systems with limited radio resources such as constraint on the training latency and constraints on the transmission power and bandwidth, transmission of a large number of model parameters inevitably suffers from quantization errors (QE) and transmission outage (TO). In this paper, we consider such non-ideal wireless channels, and carry out the first analysis showing that the FL convergence can be severely jeopardized by TO and QE, but intriguingly can be alleviated if the clients have uniform outage probabilities. These insightful results motivate us to propose a robust FL scheme, named FedTOE, which performs joint allocation of wireless resources and quantization bits across the clients to minimize the QE while making the clients have the same TO probability. Extensive experimental results are presented to show the superior performance of FedTOE for a deep learning-based classification task with transmission latency constraints.

A Self-supervised Method for Entity Alignment arxiv:2106.09395 📈 6

Xiao Liu, Haoyun Hong, Xinghao Wang, Zeyi Chen, Evgeny Kharlamov, Yuxiao Dong, Jie Tang

**Abstract:** Entity alignment, aiming to identify equivalent entities across different knowledge graphs (KGs), is a fundamental problem for constructing large-scale KGs. Over the course of its development, supervision has been considered necessary for accurate alignments. Inspired by the recent progress of self-supervised learning, we explore the extent to which we can get rid of supervision for entity alignment. Existing supervised methods for this task focus on pulling each pair of positive (labeled) entities close to each other. However, our analysis suggests that the learning of entity alignment can actually benefit more from pushing sampled (unlabeled) negatives far away than pulling positive aligned pairs close. We present SelfKG by leveraging this discovery to design a contrastive learning strategy across two KGs. Extensive experiments on benchmark datasets demonstrate that SelfKG without supervision can match or achieve comparable results with state-of-the-art supervised baselines. The performance of SelfKG demonstrates self-supervised learning offers great potential for entity alignment in KGs.

Unsupervised Path Representation Learning with Curriculum Negative Sampling arxiv:2106.09373 📈 6

Sean Bin Yang, Chenjuan Guo, Jilin Hu, Jian Tang, Bin Yang

**Abstract:** Path representations are critical in a variety of transportation applications, such as estimating path ranking in path recommendation systems and estimating path travel time in navigation systems. Existing studies often learn task-specific path representations in a supervised manner, which require a large amount of labeled training data and generalize poorly to other tasks. We propose an unsupervised learning framework Path InfoMax (PIM) to learn generic path representations that work for different downstream tasks. We first propose a curriculum negative sampling method, for each input path, to generate a small amount of negative paths, by following the principles of curriculum learning. Next, \emph{PIM} employs mutual information maximization to learn path representations from both a global and a local view. In the global view, PIM distinguishes the representations of the input paths from those of the negative paths. In the local view, \emph{PIM} distinguishes the input path representations from the representations of the nodes that appear only in the negative paths. This enables the learned path representations to encode both global and local information at different scales. Extensive experiments on two downstream tasks, ranking score estimation and travel time estimation, using two road network datasets suggest that PIM significantly outperforms other unsupervised methods and is also able to be used as a pre-training method to enhance supervised path representation learning.

A Multi-task convolutional neural network for blind stereoscopic image quality assessment using naturalness analysis arxiv:2106.09303 📈 6

Salima Bourbia, Ayoub Karine, Aladine Chetouani, Mohammed El Hassouni

**Abstract:** This paper addresses the problem of blind stereoscopic image quality assessment (NR-SIQA) using a new multi-task deep learning based-method. In the field of stereoscopic vision, the information is fairly distributed between the left and right views as well as the binocular phenomenon. In this work, we propose to integrate these characteristics to estimate the quality of stereoscopic images without reference through a convolutional neural network. Our method is based on two main tasks: the first task predicts naturalness analysis based features adapted to stereo images, while the second task predicts the quality of such images. The former, so-called auxiliary task, aims to find more robust and relevant features to improve the quality prediction. To do this, we compute naturalness-based features using a Natural Scene Statistics (NSS) model in the complex wavelet domain. It allows to capture the statistical dependency between pairs of the stereoscopic images. Experiments are conducted on the well known LIVE PHASE I and LIVE PHASE II databases. The results obtained show the relevance of our method when comparing with those of the state-of-the-art. Our code is available online on https://github.com/Bourbia-Salima/multitask-cnn-nrsiqa_2021.

Message Passing in Graph Convolution Networks via Adaptive Filter Banks arxiv:2106.09910 📈 5

Xing Gao, Wenrui Dai, Chenglin Li, Junni Zou, Hongkai Xiong, Pascal Frossard

**Abstract:** Graph convolution networks, like message passing graph convolution networks (MPGCNs), have been a powerful tool in representation learning of networked data. However, when data is heterogeneous, most architectures are limited as they employ a single strategy to handle multi-channel graph signals and they typically focus on low-frequency information. In this paper, we present a novel graph convolution operator, termed BankGCN, which keeps benefits of message passing models, but extends their capabilities beyond `low-pass' features. It decomposes multi-channel signals on graphs into subspaces and handles particular information in each subspace with an adapted filter. The filters of all subspaces have different frequency responses and together form a filter bank. Furthermore, each filter in the spectral domain corresponds to a message passing scheme, and diverse schemes are implemented via the filter bank. Importantly, the filter bank and the signal decomposition are jointly learned to adapt to the spectral characteristics of data and to target applications. Furthermore, this is implemented almost without extra parameters in comparison with most existing MPGCNs. Experimental results show that the proposed convolution operator permits to achieve excellent performance in graph classification on a collection of benchmark graph datasets.

GEM: A General Evaluation Benchmark for Multimodal Tasks arxiv:2106.09889 📈 5

Lin Su, Nan Duan, Edward Cui, Lei Ji, Chenfei Wu, Huaishao Luo, Yongfei Liu, Ming Zhong, Taroon Bharti, Arun Sacheti

**Abstract:** In this paper, we present GEM as a General Evaluation benchmark for Multimodal tasks. Different from existing datasets such as GLUE, SuperGLUE, XGLUE and XTREME that mainly focus on natural language tasks, GEM is a large-scale vision-language benchmark, which consists of GEM-I for image-language tasks and GEM-V for video-language tasks. Comparing with existing multimodal datasets such as MSCOCO and Flicker30K for image-language tasks, YouCook2 and MSR-VTT for video-language tasks, GEM is not only the largest vision-language dataset covering image-language tasks and video-language tasks at the same time, but also labeled in multiple languages. We also provide two baseline models for this benchmark. We will release the dataset, code and baseline models, aiming to advance the development of multilingual multimodal research.

An Improved Single Step Non-autoregressive Transformer for Automatic Speech Recognition arxiv:2106.09885 📈 5

Ruchao Fan, Wei Chu, Peng Chang, Jing Xiao, Abeer Alwan

**Abstract:** Non-autoregressive mechanisms can significantly decrease inference time for speech transformers, especially when the single step variant is applied. Previous work on CTC alignment-based single step non-autoregressive transformer (CASS-NAT) has shown a large real time factor (RTF) improvement over autoregressive transformers (AT). In this work, we propose several methods to improve the accuracy of the end-to-end CASS-NAT, followed by performance analyses. First, convolution augmented self-attention blocks are applied to both the encoder and decoder modules. Second, we propose to expand the trigger mask (acoustic boundary) for each token to increase the robustness of CTC alignments. In addition, iterated loss functions are used to enhance the gradient update of low-layer parameters. Without using an external language model, the WERs of the improved CASS-NAT, when using the three methods, are 3.1%/7.2% on Librispeech test clean/other sets and the CER is 5.4% on the Aishell1 test set, achieving a 7%~21% relative WER/CER improvement. For the analyses, we plot attention weight distributions in the decoders to visualize the relationships between token-level acoustic embeddings. When the acoustic embeddings are visualized, we find that they have a similar behavior to word embeddings, which explains why the improved CASS-NAT performs similarly to AT.

Heuristic Stopping Rules For Technology-Assisted Review arxiv:2106.09871 📈 5

Eugene Yang, David D. Lewis, Ophir Frieder

**Abstract:** Technology-assisted review (TAR) refers to human-in-the-loop active learning workflows for finding relevant documents in large collections. These workflows often must meet a target for the proportion of relevant documents found (i.e. recall) while also holding down costs. A variety of heuristic stopping rules have been suggested for striking this tradeoff in particular settings, but none have been tested against a range of recall targets and tasks. We propose two new heuristic stopping rules, Quant and QuantCI based on model-based estimation techniques from survey research. We compare them against a range of proposed heuristics and find they are accurate at hitting a range of recall targets while substantially reducing review costs.

On Invariance Penalties for Risk Minimization arxiv:2106.09777 📈 5

Kia Khezeli, Arno Blaas, Frank Soboczenski, Nicholas Chia, John Kalantari

**Abstract:** The Invariant Risk Minimization (IRM) principle was first proposed by Arjovsky et al. [2019] to address the domain generalization problem by leveraging data heterogeneity from differing experimental conditions. Specifically, IRM seeks to find a data representation under which an optimal classifier remains invariant across all domains. Despite the conceptual appeal of IRM, the effectiveness of the originally proposed invariance penalty has recently been brought into question. In particular, there exists counterexamples for which that invariance penalty can be arbitrarily small for non-invariant data representations. We propose an alternative invariance penalty by revisiting the Gramian matrix of the data representation. We discuss the role of its eigenvalues in the relationship between the risk and the invariance penalty, and demonstrate that it is ill-conditioned for said counterexamples. The proposed approach is guaranteed to recover an invariant representation for linear settings under mild non-degeneracy conditions. Its effectiveness is substantiated by experiments on DomainBed and InvarianceUnitTest, two extensive test beds for domain generalization.

Synthetic COVID-19 Chest X-ray Dataset for Computer-Aided Diagnosis arxiv:2106.09759 📈 5

Hasib Zunair, A. Ben Hamza

**Abstract:** We introduce a new dataset called Synthetic COVID-19 Chest X-ray Dataset for training machine learning models. The dataset consists of 21,295 synthetic COVID-19 chest X-ray images to be used for computer-aided diagnosis. These images, generated via an unsupervised domain adaptation approach, are of high quality. We find that the synthetic images not only improve performance of various deep learning architectures when used as additional training data under heavy imbalance conditions, but also detect the target class with high confidence. We also find that comparable performance can also be achieved when trained only on synthetic images. Further, salient features of the synthetic COVID-19 images indicate that the distribution is significantly different from Non-COVID-19 classes, enabling a proper decision boundary. We hope the availability of such high fidelity chest X-ray images of COVID-19 will encourage advances in the development of diagnostic and/or management tools.

Privacy-Preserving Eye-tracking Using Deep Learning arxiv:2106.09621 📈 5

Salman Seyedi, Zifan Jiang, Allan Levey, Gari D. Clifford

**Abstract:** The expanding usage of complex machine learning methods like deep learning has led to an explosion in human activity recognition, particularly applied to health. In particular, as part of a larger body sensor network system, face and full-body analysis is becoming increasingly common for evaluating health status. However, complex models which handle private and sometimes protected data, raise concerns about the potential leak of identifiable data. In this work, we focus on the case of a deep network model trained on images of individual faces. Full-face video recordings taken from 493 individuals undergoing an eye-tracking based evaluation of neurological function were used. Outputs, gradients, intermediate layer outputs, loss, and labels were used as inputs for a deep network with an added support vector machine emission layer to recognize membership in the training data. The inference attack method and associated mathematical analysis indicate that there is a low likelihood of unintended memorization of facial features in the deep learning model. In this study, it is showed that the named model preserves the integrity of training data with reasonable confidence. The same process can be implemented in similar conditions for different models.

Disentangling Identifiable Features from Noisy Data with Structured Nonlinear ICA arxiv:2106.09620 📈 5

Hermanni Hälvä, Sylvain Le Corff, Luc Lehéricy, Jonathan So, Yongjie Zhu, Elisabeth Gassiat, Aapo Hyvarinen

**Abstract:** We introduce a new general identifiable framework for principled disentanglement referred to as Structured Nonlinear Independent Component Analysis (SNICA). Our contribution is to extend the identifiability theory of deep generative models for a very broad class of structured models. While previous works have shown identifiability for specific classes of time-series models, our theorems extend this to more general temporal structures as well as to models with more complex structures such as spatial dependencies. In particular, we establish the major result that identifiability for this framework holds even in the presence of noise of unknown distribution. Finally, as an example of our framework's flexibility, we introduce the first nonlinear ICA model for time-series that combines the following very useful properties: it accounts for both nonstationarity and autocorrelation in a fully unsupervised setting; performs dimensionality reduction; models hidden states; and enables principled estimation and inference by variational maximum-likelihood.

Optimality and Stability in Federated Learning: A Game-theoretic Approach arxiv:2106.09580 📈 5

Kate Donahue, Jon Kleinberg

**Abstract:** Federated learning is a distributed learning paradigm where multiple agents, each only with access to local data, jointly learn a global model. There has recently been an explosion of research aiming not only to improve the accuracy rates of federated learning, but also provide certain guarantees around social good properties such as total error. One branch of this research has taken a game-theoretic approach, and in particular, prior work has viewed federated learning as a hedonic game, where error-minimizing players arrange themselves into federating coalitions. This past work proves the existence of stable coalition partitions, but leaves open a wide range of questions, including how far from optimal these stable solutions are. In this work, we motivate and define a notion of optimality given by the average error rates among federating agents (players). First, we provide and prove the correctness of an efficient algorithm to calculate an optimal (error minimizing) arrangement of players. Next, we analyze the relationship between the stability and optimality of an arrangement. First, we show that for some regions of parameter space, all stable arrangements are optimal (Price of Anarchy equal to 1). However, we show this is not true for all settings: there exist examples of stable arrangements with higher cost than optimal (Price of Anarchy greater than 1). Finally, we give the first constant-factor bound on the performance gap between stability and optimality, proving that the total error of the worst stable solution can be no higher than 9 times the total error of an optimal solution (Price of Anarchy bound of 9).

Adversarial Visual Robustness by Causal Intervention arxiv:2106.09534 📈 5

Kaihua Tang, Mingyuan Tao, Hanwang Zhang

**Abstract:** Adversarial training is the de facto most promising defense against adversarial examples. Yet, its passive nature inevitably prevents it from being immune to unknown attackers. To achieve a proactive defense, we need a more fundamental understanding of adversarial examples, beyond the popular bounded threat model. In this paper, we provide a causal viewpoint of adversarial vulnerability: the cause is the spurious correlation ubiquitously existing in learning, i.e., the confounding effect, where attackers are precisely exploiting these effects. Therefore, a fundamental solution for adversarial robustness is by causal intervention. As these visual confounders are imperceptible in general, we propose to use the instrumental variable that achieves causal intervention without the need for confounder observation. We term our robust training method as Causal intervention by instrumental Variable (CiiV). It's a causal regularization that 1) augments the image with multiple retinotopic centers and 2) encourages the model to learn causal features, rather than local confounding patterns, by favoring features linearly responding to spatial interpolations. Extensive experiments on a wide spectrum of attackers and settings applied in CIFAR-10, CIFAR-100, and mini-ImageNet demonstrate that CiiV is robust to adaptive attacks, including the recent AutoAttack. Besides, as a general causal regularization, it can be easily plugged into other methods to further boost the robustness.

Machine learning methods for postprocessing ensemble forecasts of wind gusts: A systematic comparison arxiv:2106.09512 📈 5

Benedikt Schulz, Sebastian Lerch

**Abstract:** Postprocessing ensemble weather predictions to correct systematic errors has become a standard practice in research and operations. However, only few recent studies have focused on ensemble postprocessing of wind gust forecasts, despite its importance for severe weather warnings. Here, we provide a comprehensive review and systematic comparison of eight statistical and machine learning methods for probabilistic wind gust forecasting via ensemble postprocessing, that can be divided in three groups: State of the art postprocessing techniques from statistics (ensemble model output statistics (EMOS), member-by-member postprocessing, isotonic distributional regression), established machine learning methods (gradient-boosting extended EMOS, quantile regression forests) and neural network-based approaches (distributional regression network, Bernstein quantile network, histogram estimation network). The methods are systematically compared using six years of data from a high-resolution, convection-permitting ensemble prediction system that was run operationally at the German weather service, and hourly observations at 175 surface weather stations in Germany. While all postprocessing methods yield calibrated forecasts and are able to correct the systematic errors of the raw ensemble predictions, incorporating information from additional meteorological predictor variables beyond wind gusts leads to significant improvements in forecast skill. In particular, we propose a flexible framework of locally adaptive neural networks with different probabilistic forecast types as output, which not only significantly outperform all benchmark postprocessing methods but also learn physically consistent relations associated with the diurnal cycle, especially the evening transition of the planetary boundary layer.

Secure Multi-Function Computation with Private Remote Sources arxiv:2106.09485 📈 5

Onur Günlü, Matthieu Bloch, Rafael F. Schaefer

**Abstract:** We consider a distributed function computation problem in which parties observing noisy versions of a remote source facilitate the computation of a function of their observations at a fusion center through public communication. The distributed function computation is subject to constraints, including not only reliability and storage but also privacy and secrecy. Specifically, 1) the remote source should remain private from an eavesdropper and the fusion center, measured in terms of the information leaked about the remote source; 2) the function computed should remain secret from the eavesdropper, measured in terms of the information leaked about the arguments of the function, to ensure secrecy regardless of the exact function used. We derive the exact rate regions for lossless and lossy single-function computation and illustrate the lossy single-function computation rate region for an information bottleneck example, in which the optimal auxiliary random variables are characterized for binary-input symmetric-output channels. We extend the approach to lossless and lossy asynchronous multiple-function computations with joint secrecy and privacy constraints, in which case inner and outer bounds for the rate regions differing only in the Markov chain conditions imposed are characterized.

Modelling resource allocation in uncertain system environment through deep reinforcement learning arxiv:2106.09461 📈 5

Neel Gandhi, Shakti Mishra

**Abstract:** Reinforcement Learning has applications in field of mechatronics, robotics, and other resource-constrained control system. Problem of resource allocation is primarily solved using traditional predefined techniques and modern deep learning methods. The drawback of predefined and most deep learning methods for resource allocation is failing to meet the requirements in cases of uncertain system environment. We can approach problem of resource allocation in uncertain system environment alongside following certain criteria using deep reinforcement learning. Also, reinforcement learning has ability for adapting to new uncertain environment for prolonged period of time. The paper provides a detailed comparative analysis on various deep reinforcement learning methods by applying different components to modify architecture of reinforcement learning with use of noisy layers, prioritized replay, bagging, duelling networks, and other related combination to obtain improvement in terms of performance and reduction of computational cost. The paper identifies problem of resource allocation in uncertain environment could be effectively solved using Noisy Bagging duelling double deep Q network achieving efficiency of 97.7% by maximizing reward with significant exploration in given simulated environment for resource allocation.

Unsupervised Training Data Generation of Handwritten Formulas using Generative Adversarial Networks with Self-Attention arxiv:2106.09432 📈 5

Matthias Springstein, Eric Müller-Budack, Ralph Ewerth

**Abstract:** The recognition of handwritten mathematical expressions in images and video frames is a difficult and unsolved problem yet. Deep convectional neural networks are basically a promising approach, but typically require a large amount of labeled training data. However, such a large training dataset does not exist for the task of handwritten formula recognition. In this paper, we introduce a system that creates a large set of synthesized training examples of mathematical expressions which are derived from LaTeX documents. For this purpose, we propose a novel attention-based generative adversarial network to translate rendered equations to handwritten formulas. The datasets generated by this approach contain hundreds of thousands of formulas, making it ideal for pretraining or the design of more complex models. We evaluate our synthesized dataset and the recognition approach on the CROHME 2014 benchmark dataset. Experimental results demonstrate the feasibility of the approach.

Interpretable Machine Learning Classifiers for Brain Tumour Survival Prediction arxiv:2106.09424 📈 5

Colleen E. Charlton, Michael Tin Chung Poon, Paul M. Brennan, Jacques D. Fleuriot

**Abstract:** Prediction of survival in patients diagnosed with a brain tumour is challenging because of heterogeneous tumour behaviours and responses to treatment. Better estimations of prognosis would support treatment planning and patient support. Advances in machine learning have informed development of clinical predictive models, but their integration into clinical practice is almost non-existent. One reasons for this is the lack of interpretability of models. In this paper, we use a novel brain tumour dataset to compare two interpretable rule list models against popular machine learning approaches for brain tumour survival prediction. All models are quantitatively evaluated using standard performance metrics. The rule lists are also qualitatively assessed for their interpretability and clinical utility. The interpretability of the black box machine learning models is evaluated using two post-hoc explanation techniques, LIME and SHAP. Our results show that the rule lists were only slightly outperformed by the black box models. We demonstrate that rule list algorithms produced simple decision lists that align with clinical expertise. By comparison, post-hoc interpretability methods applied to black box models may produce unreliable explanations of local model predictions. Model interpretability is essential for understanding differences in predictive performance and for integration into clinical practice.

A Simple Generative Network arxiv:2106.09330 📈 5

Daniel N. Nissani

**Abstract:** Generative neural networks are able to mimic intricate probability distributions such as those of handwritten text, natural images, etc. Since their inception several models were proposed. The most successful of these were based on adversarial (GAN), auto-encoding (VAE) and maximum mean discrepancy (MMD) relatively complex architectures and schemes. Surprisingly, a very simple architecture (a single feed-forward neural network) in conjunction with an obvious optimization goal (Kullback_Leibler divergence) was apparently overlooked. This paper demonstrates that such a model (denoted SGN for its simplicity) is able to generate samples visually and quantitatively competitive as compared with the fore-mentioned state of the art methods.

Towards Understanding Deep Learning from Noisy Labels with Small-Loss Criterion arxiv:2106.09291 📈 5

Xian-Jin Gui, Wei Wang, Zhang-Hao Tian

**Abstract:** Deep neural networks need large amounts of labeled data to achieve good performance. In real-world applications, labels are usually collected from non-experts such as crowdsourcing to save cost and thus are noisy. In the past few years, deep learning methods for dealing with noisy labels have been developed, many of which are based on the small-loss criterion. However, there are few theoretical analyses to explain why these methods could learn well from noisy labels. In this paper, we theoretically explain why the widely-used small-loss criterion works. Based on the explanation, we reformalize the vanilla small-loss criterion to better tackle noisy labels. The experimental results verify our theoretical explanation and also demonstrate the effectiveness of the reformalization.

A Random CNN Sees Objects: One Inductive Bias of CNN and Its Applications arxiv:2106.09259 📈 5

Yun-Hao Cao, Jianxin Wu

**Abstract:** This paper starts by revealing a surprising finding: without any learning, a randomly initialized CNN can localize objects surprisingly well. That is, a CNN has an inductive bias to naturally focus on objects, named as Tobias ("The object is at sight") in this paper. This empirical inductive bias is further analyzed and successfully applied to self-supervised learning (SSL). A CNN is encouraged to learn representations that focus on the foreground object, by transforming every image into various versions with different backgrounds, where the foreground and background separation is guided by Tobias. Experimental results show that the proposed Tobias significantly improves downstream tasks, especially for object detection. This paper also shows that Tobias has consistent improvements on training sets of different sizes, and is more resilient to changes in image augmentations. Code is available at https://github.com/CupidJay/Tobias.

Federated CycleGAN for Privacy-Preserving Image-to-Image Translation arxiv:2106.09246 📈 5

Joonyoung Song, Jong Chul Ye

**Abstract:** Unsupervised image-to-image translation methods such as CycleGAN learn to convert images from one domain to another using unpaired training data sets from different domains. Unfortunately, these approaches still require centrally collected unpaired records, potentially violating privacy and security issues. Although the recent federated learning (FL) allows a neural network to be trained without data exchange, the basic assumption of the FL is that all clients have their own training data from a similar domain, which is different from our image-to-image translation scenario in which each client has images from its unique domain and the goal is to learn image translation between different domains without accessing the target domain data. To address this, here we propose a novel federated CycleGAN architecture that can learn image translation in an unsupervised manner while maintaining the data privacy. Specifically, our approach arises from a novel observation that CycleGAN loss can be decomposed into the sum of client specific local objectives that can be evaluated using only their data. This local objective decomposition allows multiple clients to participate in federated CycleGAN training without sacrificing performance. Furthermore, our method employs novel switchable generator and discriminator architecture using Adaptive Instance Normalization (AdaIN) that significantly reduces the band-width requirement of the federated learning. Our experimental results on various unsupervised image translation tasks show that our federated CycleGAN provides comparable performance compared to the non-federated counterpart.

SE-MD: A Single-encoder multiple-decoder deep network for point cloud generation from 2D images arxiv:2106.15325 📈 4

Abdul Mueed Hafiz, Rouf Ul Alam Bhat, Shabir Ahmad Parah, M. Hassaballah

**Abstract:** 3D model generation from single 2D RGB images is a challenging and actively researched computer vision task. Various techniques using conventional network architectures have been proposed for the same. However, the body of research work is limited and there are various issues like using inefficient 3D representation formats, weak 3D model generation backbones, inability to generate dense point clouds, dependence of post-processing for generation of dense point clouds, and dependence on silhouettes in RGB images. In this paper, a novel 2D RGB image to point cloud conversion technique is proposed, which improves the state of art in the field due to its efficient, robust and simple model by using the concept of parallelization in network architecture. It not only uses the efficient and rich 3D representation of point clouds, but also uses a novel and robust point cloud generation backbone in order to address the prevalent issues. This involves using a single-encoder multiple-decoder deep network architecture wherein each decoder generates certain fixed viewpoints. This is followed by fusing all the viewpoints to generate a dense point cloud. Various experiments are conducted on the technique and its performance is compared with those of other state of the art techniques and impressive gains in performance are demonstrated. Code is available at https://github.com/mueedhafiz1982/

Effective Model Sparsification by Scheduled Grow-and-Prune Methods arxiv:2106.09857 📈 4

Xiaolong Ma, Minghai Qin, Fei Sun, Zejiang Hou, Kun Yuan, Yi Xu, Yanzhi Wang, Yen-Kuang Chen, Rong Jin, Yuan Xie

**Abstract:** Deep neural networks (DNNs) are effective in solving many real-world problems. Larger DNN models usually exhibit better quality (e.g., accuracy) but their excessive computation results in long inference time. Model sparsification can reduce the computation and memory cost while maintaining model quality. Most existing sparsification algorithms unidirectionally remove weights, while others randomly or greedily explore a small subset of weights in each layer for pruning. The limitations of these algorithms reduce the level of achievable sparsity. In addition, many algorithms still require pre-trained dense models and thus suffer from large memory footprint. In this paper, we propose a novel scheduled grow-and-prune (GaP) methodology without having to pre-train a dense model. It addresses the shortcomings of the previous works by repeatedly growing a subset of layers to dense and then pruning them back to sparse after some training. Experiments show that the models pruned using the proposed methods match or beat the quality of the highly optimized dense models at 80% sparsity on a variety of tasks, such as image classification, objective detection, 3D object part segmentation, and translation. They also outperform other state-of-the-art (SOTA) methods for model sparsification. As an example, a 90% non-uniform sparse ResNet-50 model obtained via GaP achieves 77.9% top-1 accuracy on ImageNet, improving the previous SOTA results by 1.5%. All code will be publicly released.

A Distance-based Separability Measure for Internal Cluster Validation arxiv:2106.09794 📈 4

Shuyue Guan, Murray Loew

**Abstract:** To evaluate clustering results is a significant part of cluster analysis. Since there are no true class labels for clustering in typical unsupervised learning, many internal cluster validity indices (CVIs), which use predicted labels and data, have been created. Without true labels, to design an effective CVI is as difficult as to create a clustering method. And it is crucial to have more CVIs because there are no universal CVIs that can be used to measure all datasets and no specific methods of selecting a proper CVI for clusters without true labels. Therefore, to apply a variety of CVIs to evaluate clustering results is necessary. In this paper, we propose a novel internal CVI -- the Distance-based Separability Index (DSI), based on a data separability measure. We compared the DSI with eight internal CVIs including studies from early Dunn (1974) to most recent CVDD (2019) and an external CVI as ground truth, by using clustering results of five clustering algorithms on 12 real and 97 synthetic datasets. Results show DSI is an effective, unique, and competitive CVI to other compared CVIs. We also summarized the general process to evaluate CVIs and created the rank-difference metric for comparison of CVIs' results.

Hi-Phy: A Benchmark for Hierarchical Physical Reasoning arxiv:2106.09692 📈 4

Cheng Xue, Vimukthini Pinto, Chathura Gamage, Peng Zhang, Jochen Renz

**Abstract:** Reasoning about the behaviour of physical objects is a key capability of agents operating in physical worlds. Humans are very experienced in physical reasoning while it remains a major challenge for AI. To facilitate research addressing this problem, several benchmarks have been proposed recently. However, these benchmarks do not enable us to measure an agent's granular physical reasoning capabilities when solving a complex reasoning task. In this paper, we propose a new benchmark for physical reasoning that allows us to test individual physical reasoning capabilities. Inspired by how humans acquire these capabilities, we propose a general hierarchy of physical reasoning capabilities with increasing complexity. Our benchmark tests capabilities according to this hierarchy through generated physical reasoning tasks in the video game Angry Birds. This benchmark enables us to conduct a comprehensive agent evaluation by measuring the agent's granular physical reasoning capabilities. We conduct an evaluation with human players, learning agents, and heuristic agents and determine their capabilities. Our evaluation shows that learning agents, with good local generalization ability, still struggle to learn the underlying physical reasoning capabilities and perform worse than current state-of-the-art heuristic agents and humans. We believe that this benchmark will encourage researchers to develop intelligent agents with advanced, human-like physical reasoning capabilities. URL: https://github.com/Cheng-Xue/Hi-Phy

Design and Analysis of Robust Deep Learning Models for Stock Price Prediction arxiv:2106.09664 📈 4

Jaydip Sen, Sidra Mehtab

**Abstract:** Building predictive models for robust and accurate prediction of stock prices and stock price movement is a challenging research problem to solve. The well-known efficient market hypothesis believes in the impossibility of accurate prediction of future stock prices in an efficient stock market as the stock prices are assumed to be purely stochastic. However, numerous works proposed by researchers have demonstrated that it is possible to predict future stock prices with a high level of precision using sophisticated algorithms, model architectures, and the selection of appropriate variables in the models. This chapter proposes a collection of predictive regression models built on deep learning architecture for robust and precise prediction of the future prices of a stock listed in the diversified sectors in the National Stock Exchange (NSE) of India. The Metastock tool is used to download the historical stock prices over a period of two years (2013- 2014) at 5 minutes intervals. While the records for the first year are used to train the models, the testing is carried out using the remaining records. The design approaches of all the models and their performance results are presented in detail. The models are also compared based on their execution time and accuracy of prediction.

A Short Note of PAGE: Optimal Convergence Rates for Nonconvex Optimization arxiv:2106.09663 📈 4

Zhize Li

**Abstract:** In this note, we first recall the nonconvex problem setting and introduce the optimal PAGE algorithm (Li et al., ICML'21). Then we provide a simple and clean convergence analysis of PAGE for achieving optimal convergence rates. Moreover, PAGE and its analysis can be easily adopted and generalized to other works. We hope that this note provides the insights and is helpful for future works.

Stochastic Bias-Reduced Gradient Methods arxiv:2106.09481 📈 4

Hilal Asi, Yair Carmon, Arun Jambulapati, Yujia Jin, Aaron Sidford

**Abstract:** We develop a new primitive for stochastic optimization: a low-bias, low-cost estimator of the minimizer $x_\star$ of any Lipschitz strongly-convex function. In particular, we use a multilevel Monte-Carlo approach due to Blanchet and Glynn to turn any optimal stochastic gradient method into an estimator of $x_\star$ with bias $δ$, variance $O(\log(1/δ))$, and an expected sampling cost of $O(\log(1/δ))$ stochastic gradient evaluations. As an immediate consequence, we obtain cheap and nearly unbiased gradient estimators for the Moreau-Yoshida envelope of any Lipschitz convex function, allowing us to perform dimension-free randomized smoothing. We demonstrate the potential of our estimator through four applications. First, we develop a method for minimizing the maximum of $N$ functions, improving on recent results and matching a lower bound up to logarithmic factors. Second and third, we recover state-of-the-art rates for projection-efficient and gradient-efficient optimization using simple algorithms with a transparent analysis. Finally, we show that an improved version of our estimator would yield a nearly linear-time, optimal-utility, differentially-private non-smooth stochastic optimization method.

Importance measures derived from random forests: characterisation and extension arxiv:2106.09473 📈 4

Antonio Sutera

**Abstract:** Nowadays new technologies, and especially artificial intelligence, are more and more established in our society. Big data analysis and machine learning, two sub-fields of artificial intelligence, are at the core of many recent breakthroughs in many application fields (e.g., medicine, communication, finance, ...), including some that are strongly related to our day-to-day life (e.g., social networks, computers, smartphones, ...). In machine learning, significant improvements are usually achieved at the price of an increasing computational complexity and thanks to bigger datasets. Currently, cutting-edge models built by the most advanced machine learning algorithms typically became simultaneously very efficient and profitable but also extremely complex. Their complexity is to such an extent that these models are commonly seen as black-boxes providing a prediction or a decision which can not be interpreted or justified. Nevertheless, whether these models are used autonomously or as a simple decision-making support tool, they are already being used in machine learning applications where health and human life are at stake. Therefore, it appears to be an obvious necessity not to blindly believe everything coming out of those models without a detailed understanding of their predictions or decisions. Accordingly, this thesis aims at improving the interpretability of models built by a specific family of machine learning algorithms, the so-called tree-based methods. Several mechanisms have been proposed to interpret these models and we aim along this thesis to improve their understanding, study their properties, and define their limitations.

Towards Heterogeneous Clients with Elastic Federated Learning arxiv:2106.09433 📈 4

Zichen Ma, Yu Lu, Zihan Lu, Wenye Li, Jinfeng Yi, Shuguang Cui

**Abstract:** Federated learning involves training machine learning models over devices or data silos, such as edge processors or data warehouses, while keeping the data local. Training in heterogeneous and potentially massive networks introduces bias into the system, which is originated from the non-IID data and the low participation rate in reality. In this paper, we propose Elastic Federated Learning (EFL), an unbiased algorithm to tackle the heterogeneity in the system, which makes the most informative parameters less volatile during training, and utilizes the incomplete local updates. It is an efficient and effective algorithm that compresses both upstream and downstream communications. Theoretically, the algorithm has convergence guarantee when training on the non-IID data at the low participation rate. Empirical experiments corroborate the competitive performance of EFL framework on the robustness and the efficiency.

Identifiability of AMP chain graph models arxiv:2106.09350 📈 4

Yuhao Wang, Arnab Bhattacharyya

**Abstract:** We study identifiability of Andersson-Madigan-Perlman (AMP) chain graph models, which are a common generalization of linear structural equation models and Gaussian graphical models. AMP models are described by DAGs on chain components which themselves are undirected graphs. For a known chain component decomposition, we show that the DAG on the chain components is identifiable if the determinants of the residual covariance matrices of the chain components are monotone non-decreasing in topological order. This condition extends the equal variance identifiability criterion for Bayes nets, and it can be generalized from determinants to any super-additive function on positive semidefinite matrices. When the component decomposition is unknown, we describe conditions that allow recovery of the full structure using a polynomial time algorithm based on submodular function minimization. We also conduct experiments comparing our algorithm's performance against existing baselines.

MHNF: Multi-hop Heterogeneous Neighborhood information Fusion graph representation learning arxiv:2106.09289 📈 4

Dongjie Zhu, Yundong Sun, Haiwen Du, Zhaoshuo Tian

**Abstract:** Attention mechanism enables the Graph Neural Networks(GNNs) to learn the attention weights between the target node and its one-hop neighbors, the performance is further improved. However, the most existing GNNs are oriented to homogeneous graphs and each layer can only aggregate the information of one-hop neighbors. Stacking multi-layer networks will introduce a lot of noise and easily lead to over smoothing. We propose a Multi-hop Heterogeneous Neighborhood information Fusion graph representation learning method (MHNF). Specifically, we first propose a hybrid metapath autonomous extraction model to efficiently extract multi-hop hybrid neighbors. Then, we propose a hop-level heterogeneous Information aggregation model, which selectively aggregates different-hop neighborhood information within the same hybrid metapath. Finally, a hierarchical semantic attention fusion model (HSAF) is proposed, which can efficiently integrate different-hop and different-path neighborhood information respectively. This paper can solve the problem of aggregating the multi-hop neighborhood information and can learn hybrid metapaths for target task, reducing the limitation of manually specifying metapaths. In addition, HSAF can extract the internal node information of the metapaths and better integrate the semantic information of different levels. Experimental results on real datasets show that MHNF is superior to state-of-the-art methods in node classification and clustering tasks (10.94% - 69.09% and 11.58% - 394.93% relative improvement on average, respectively).

Batch Multi-Fidelity Bayesian Optimization with Deep Auto-Regressive Networks arxiv:2106.09884 📈 3

Shibo Li, Robert M. Kirby, Shandian Zhe

**Abstract:** Bayesian optimization (BO) is a powerful approach for optimizing black-box, expensive-to-evaluate functions. To enable a flexible trade-off between the cost and accuracy, many applications allow the function to be evaluated at different fidelities. In order to reduce the optimization cost while maximizing the benefit-cost ratio, in this paper, we propose Batch Multi-fidelity Bayesian Optimization with Deep Auto-Regressive Networks (BMBO-DARN). We use a set of Bayesian neural networks to construct a fully auto-regressive model, which is expressive enough to capture strong yet complex relationships across all the fidelities, so as to improve the surrogate learning and optimization performance. Furthermore, to enhance the quality and diversity of queries, we develop a simple yet efficient batch querying method, without any combinatorial search over the fidelities. We propose a batch acquisition function based on Max-value Entropy Search (MES) principle, which penalizes highly correlated queries and encourages diversity. We use posterior samples and moment matching to fulfill efficient computation of the acquisition function and conduct alternating optimization over every fidelity-input pair, which guarantees an improvement at each step. We demonstrate the advantage of our approach on four real-world hyperparameter optimization applications.

Smoothed Multi-View Subspace Clustering arxiv:2106.09875 📈 3

Peng Chen, Liang Liu, Zhengrui Ma, Zhao Kang

**Abstract:** In recent years, multi-view subspace clustering has achieved impressive performance due to the exploitation of complementary imformation across multiple views. However, multi-view data can be very complicated and are not easy to cluster in real-world applications. Most existing methods operate on raw data and may not obtain the optimal solution. In this work, we propose a novel multi-view clustering method named smoothed multi-view subspace clustering (SMVSC) by employing a novel technique, i.e., graph filtering, to obtain a smooth representation for each view, in which similar data points have similar feature values. Specifically, it retains the graph geometric features through applying a low-pass filter. Consequently, it produces a ``clustering-friendly" representation and greatly facilitates the downstream clustering task. Extensive experiments on benchmark datasets validate the superiority of our approach. Analysis shows that graph filtering increases the separability of classes.

Towards Clustering-friendly Representations: Subspace Clustering via Graph Filtering arxiv:2106.09874 📈 3

Zhengrui Ma, Zhao Kang, Guangchun Luo, Ling Tian

**Abstract:** Finding a suitable data representation for a specific task has been shown to be crucial in many applications. The success of subspace clustering depends on the assumption that the data can be separated into different subspaces. However, this simple assumption does not always hold since the raw data might not be separable into subspaces. To recover the ``clustering-friendly'' representation and facilitate the subsequent clustering, we propose a graph filtering approach by which a smooth representation is achieved. Specifically, it injects graph similarity into data features by applying a low-pass filter to extract useful data representations for clustering. Extensive experiments on image and document clustering datasets demonstrate that our method improves upon state-of-the-art subspace clustering techniques. Especially, its comparable performance with deep learning methods emphasizes the effectiveness of the simple graph filtering scheme for many real-world applications. An ablation study shows that graph filtering can remove noise, preserve structure in the image, and increase the separability of classes.

PAC Prediction Sets Under Covariate Shift arxiv:2106.09848 📈 3

Sangdon Park, Edgar Dobriban, Insup Lee, Osbert Bastani

**Abstract:** An important challenge facing modern machine learning is how to rigorously quantify the uncertainty of model predictions. Conveying uncertainty is especially important when there are changes to the underlying data distribution that might invalidate the predictive model. Yet, most existing uncertainty quantification algorithms break down in the presence of such shifts. We propose a novel approach that addresses this challenge by constructing \emph{probably approximately correct (PAC)} prediction sets in the presence of covariate shift. Our approach focuses on the setting where there is a covariate shift from the source distribution (where we have labeled training examples) to the target distribution (for which we want to quantify uncertainty). Our algorithm assumes given importance weights that encode how the probabilities of the training examples change under the covariate shift. In practice, importance weights typically need to be estimated; thus, we extend our algorithm to the setting where we are given confidence intervals for the importance weights rather than their true value. We demonstrate the effectiveness of our approach on various covariate shifts designed based on the DomainNet and ImageNet datasets.

Automatic Segmentation of the Prostate on 3D Trans-rectal Ultrasound Images using Statistical Shape Models and Convolutional Neural Networks arxiv:2106.09662 📈 3

Golnoosh Samei, Davood Karimi, Claudia Kesch, Septimiu Salcudean

**Abstract:** In this work we propose to segment the prostate on a challenging dataset of trans-rectal ultrasound (TRUS) images using convolutional neural networks (CNNs) and statistical shape models (SSMs). TRUS is commonly used for a number of image-guided interventions on the prostate. Fast and accurate segmentation on the organ in these images is crucial to planning and fusion with other modalities such as magnetic resonance images (MRIs) . However, TRUS has limited soft tissue contrast and signal to noise ratio which makes the task of segmenting the prostate challenging and subject to inter-observer and intra-observer variability. This is especially problematic at the base and apex where the gland boundary is hard to define. In this paper, we aim to tackle this problem by taking advantage of shape priors learnt on an MR dataset which has higher soft tissue contrast allowing the prostate to be contoured more accurately. We use this shape prior in combination with a prostate tissue probability map computed by a CNN for segmentation.

A Deep Reinforcement Learning Approach towards Pendulum Swing-up Problem based on TF-Agents arxiv:2106.09556 📈 3

Yifei Bi, Xinyi Chen, Caihui Xiao

**Abstract:** Adapting the idea of training CartPole with Deep Q-learning agent, we are able to find a promising result that prevent the pole from falling down. The capacity of reinforcement learning (RL) to learn from the interaction between the environment and agent provides an optimal control strategy. In this paper, we aim to solve the classic pendulum swing-up problem that making the learned pendulum to be in upright position and balanced. Deep Deterministic Policy Gradient algorithm is introduced to operate over continuous action domain in this problem. Salient results of optimal pendulum are proved with increasing average return, decreasing loss, and live video in the code part.

Exploring the Properties and Evolution of Neural Network Eigenspaces during Training arxiv:2106.09526 📈 3

Mats L. Richter, Leila Malihi, Anne-Kathrin Patricia Windler, Ulf Krumnack

**Abstract:** In this work we explore the information processing inside neural networks using logistic regression probes \cite{probes} and the saturation metric \cite{featurespace_saturation}. We show that problem difficulty and neural network capacity affect the predictive performance in an antagonistic manner, opening the possibility of detecting over- and under-parameterization of neural networks for a given task. We further show that the observed effects are independent from previously reported pathological patterns like the ``tail pattern'' described in \cite{featurespace_saturation}. Finally we are able to show that saturation patterns converge early during training, allowing for a quicker cycle time during analysis

Towards bio-inspired unsupervised representation learning for indoor aerial navigation arxiv:2106.09326 📈 3

Ni Wang, Ozan Catal, Tim Verbelen, Matthias Hartmann, Bart Dhoedt

**Abstract:** Aerial navigation in GPS-denied, indoor environments, is still an open challenge. Drones can perceive the environment from a richer set of viewpoints, while having more stringent compute and energy constraints than other autonomous platforms. To tackle that problem, this research displays a biologically inspired deep-learning algorithm for simultaneous localization and mapping (SLAM) and its application in a drone navigation system. We propose an unsupervised representation learning method that yields low-dimensional latent state descriptors, that mitigates the sensitivity to perceptual aliasing, and works on power-efficient, embedded hardware. The designed algorithm is evaluated on a dataset collected in an indoor warehouse environment, and initial results show the feasibility for robust indoor aerial navigation.

Seeing Differently, Acting Similarly: Imitation Learning with Heterogeneous Observations arxiv:2106.09256 📈 3

Xin-Qiang Cai, Yao-Xiang Ding, Zi-Xuan Chen, Yuan Jiang, Masashi Sugiyama, Zhi-Hua Zhou

**Abstract:** In many real-world imitation learning tasks, the demonstrator and the learner have to act in different but full observation spaces. This situation generates significant obstacles for existing imitation learning approaches to work, even when they are combined with traditional space adaptation techniques. The main challenge lies in bridging expert's occupancy measures to learner's dynamically changing occupancy measures under the different observation spaces. In this work, we model the above learning problem as Heterogeneous Observations Imitation Learning (HOIL). We propose the Importance Weighting with REjection (IWRE) algorithm based on the techniques of importance-weighting, learning with rejection, and active querying to solve the key challenge of occupancy measure matching. Experimental results show that IWRE can successfully solve HOIL tasks, including the challenging task of transforming the vision-based demonstrations to random access memory (RAM)-based policies under the Atari domain.

Pre-treatment of outliers and anomalies in plant data: Methodology and case study of a Vacuum Distillation Unit arxiv:2106.14641 📈 2

Kamil Oster, Stefan Güttel, Jonathan L. Shapiro, Lu Chen, Megan Jobson

**Abstract:** Data pre-treatment plays a significant role in improving data quality, thus allowing extraction of accurate information from raw data. One of the data pre-treatment techniques commonly used is outliers detection. The so-called 3$σ$ method is a common practice to identify the outliers. As shown in the manuscript, it does not identify all outliers, resulting in possible distortion of the overall statistics of the data. This problem can have a significant impact on further data analysis and can lead to reduction in the accuracy of predictive models. There is a plethora of various techniques for outliers detection, however, aside from theoretical work, they all require case study work. Two types of outliers were considered: short-term (erroneous data, noise) and long-term outliers (e.g. malfunctioning for longer periods). The data used were taken from the vacuum distillation unit (VDU) of an Asian refinery and included 40 physical sensors (temperature, pressure and flow rate). We used a modified method for 3$σ$ thresholds to identify the short-term outliers, i.e. ensors data are divided into chunks determined by change points and 3$σ$ thresholds are calculated within each chunk representing near-normal distribution. We have shown that piecewise 3$σ$ method offers a better approach to short-term outliers detection than 3$σ$ method applied to the entire time series. Nevertheless, this does not perform well for long-term outliers (which can represent another state in the data). In this case, we used principal component analysis (PCA) with Hotelling's $T^2$ statistics to identify the long-term outliers. The results obtained with PCA were subject to DBSCAN clustering method. The outliers (which were visually obvious and correctly detected by the PCA method) were also correctly identified by DBSCAN which supported the consistency and accuracy of the PCA method.

Development of a conversing and body temperature scanning autonomously navigating robot to help screen for COVID-19 arxiv:2106.09894 📈 2

Ryan Kim

**Abstract:** Throughout the COVID-19 pandemic, the most common symptom displayed by patients has been a fever, leading to the use of temperature scanning as a preemptive measure to detect potential carriers of the virus. Human employees with handheld thermometers have been used to fulfill this task, however this puts them at risk as they cannot be physically distanced and the sequential nature of this method leads to great inconveniences and inefficiency. The proposed solution is an autonomously navigating robot capable of conversing and scanning people's temperature to detect fevers and help screen for COVID-19. To satisfy this objective, the robot must be able to (1) navigate autonomously, (2) detect and track people, and (3) get individuals' temperature reading and converse with them if it exceeds 38°C. An autonomously navigating mobile robot is used with a manipulator controlled using a face tracking algorithm, and an end effector consisting of a thermal camera, smartphone, and chatbot. The goal is to develop a functioning solution that performs the above tasks. In addition, technical challenges encountered and their engineering solutions will be presented, and recommendations will be made for enhancements that could be incorporated when approaching commercialization.

Many Agent Reinforcement Learning Under Partial Observability arxiv:2106.09825 📈 2

Keyang He, Prashant Doshi, Bikramjit Banerjee

**Abstract:** Recent renewed interest in multi-agent reinforcement learning (MARL) has generated an impressive array of techniques that leverage deep reinforcement learning, primarily actor-critic architectures, and can be applied to a limited range of settings in terms of observability and communication. However, a continuing limitation of much of this work is the curse of dimensionality when it comes to representations based on joint actions, which grow exponentially with the number of agents. In this paper, we squarely focus on this challenge of scalability. We apply the key insight of action anonymity, which leads to permutation invariance of joint actions, to two recently presented deep MARL algorithms, MADDPG and IA2C, and compare these instantiations to another recent technique that leverages action anonymity, viz., mean-field MARL. We show that our instantiations can learn the optimal behavior in a broader class of agent networks than the mean-field method, using a recently introduced pragmatic domain.

Escaping strict saddle points of the Moreau envelope in nonsmooth optimization arxiv:2106.09815 📈 2

Damek Davis, Mateo Díaz, Dmitriy Drusvyatskiy

**Abstract:** Recent work has shown that stochastically perturbed gradient methods can efficiently escape strict saddle points of smooth functions. We extend this body of work to nonsmooth optimization, by analyzing an inexact analogue of a stochastically perturbed gradient method applied to the Moreau envelope. The main conclusion is that a variety of algorithms for nonsmooth optimization can escape strict saddle points of the Moreau envelope at a controlled rate. The main technical insight is that typical algorithms applied to the proximal subproblem yield directions that approximate the gradient of the Moreau envelope in relative terms.

Deep reinforcement learning with automated label extraction from clinical reports accurately classifies 3D MRI brain volumes arxiv:2106.09812 📈 2

Joseph Stember, Hrithwik Shalu

**Abstract:** Purpose: Image classification is perhaps the most fundamental task in imaging AI. However, labeling images is time-consuming and tedious. We have recently demonstrated that reinforcement learning (RL) can classify 2D slices of MRI brain images with high accuracy. Here we make two important steps toward speeding image classification: Firstly, we automatically extract class labels from the clinical reports. Secondly, we extend our prior 2D classification work to fully 3D image volumes from our institution. Hence, we proceed as follows: in Part 1, we extract labels from reports automatically using the SBERT natural language processing approach. Then, in Part 2, we use these labels with RL to train a classification Deep-Q Network (DQN) for 3D image volumes. Methods: For Part 1, we trained SBERT with 90 radiology report impressions. We then used the trained SBERT to predict class labels for use in Part 2. In Part 2, we applied multi-step image classification to allow for combined Deep-Q learning using 3D convolutions and TD(0) Q learning. We trained on a set of 90 images. We tested on a separate set of 61 images, again using the classes predicted from patient reports by the trained SBERT in Part 1. For comparison, we also trained and tested a supervised deep learning classification network on the same set of training and testing images using the same labels. Results: Part 1: Upon training with the corpus of radiology reports, the SBERT model had 100% accuracy for both normal and metastasis-containing scans. Part 2: Then, using these labels, whereas the supervised approach quickly overfit the training data and as expected performed poorly on the testing set (66% accuracy, just over random guessing), the reinforcement learning approach achieved an accuracy of 92%. The results were found to be statistically significant, with a p-value of 3.1 x 10^-5.

Wide stochastic networks: Gaussian limit and PAC-Bayesian training arxiv:2106.09798 📈 2

Eugenio Clerico, George Deligiannidis, Arnaud Doucet

**Abstract:** The limit of infinite width allows for substantial simplifications in the analytical study of overparameterized neural networks. With a suitable random initialization, an extremely large network is well approximated by a Gaussian process, both before and during training. In the present work, we establish a similar result for a simple stochastic architecture whose parameters are random variables. The explicit evaluation of the output distribution allows for a PAC-Bayesian training procedure that directly optimizes the generalization bound. For a large but finite-width network, we show empirically on MNIST that this training approach can outperform standard PAC-Bayesian methods.

A probabilistic database approach to autoencoder-based data cleaning arxiv:2106.09764 📈 2

R. R. Mauritz, F. P. J. Nijweide, J. Goseling, M. van Keulen

**Abstract:** Data quality problems are a large threat in data science. In this paper, we propose a data-cleaning autoencoder capable of near-automatic data quality improvement. It learns the structure and dependencies in the data and uses it as evidence to identify and correct doubtful values. We apply a probabilistic database approach to represent weak and strong evidence for attribute value repairs. A theoretical framework is provided, and experiments show that it can remove significant amounts of noise (i.e., data quality problems) from categorical and numeric probabilistic data. Our method does not require clean data. We do, however, show that manually cleaning a small fraction of the data significantly improves performance.

Optimizing robotic swarm based construction tasks arxiv:2106.09749 📈 2

Teshan Liyanage, Subha Fernando

**Abstract:** Social insects in nature such as ants, termites and bees construct their colonies collaboratively in a very efficient process. In these swarms, each insect contributes to the construction task individually showing redundant and parallel behavior of individual entities. But the robotics adaptations of these swarm's behaviors haven't yet made it to the real world at a large enough scale of commonly being used due to the limitations in the existing approaches to the swarm robotics construction. This paper presents an approach that combines the existing swarm construction approaches which results in a swarm robotic system, capable of constructing a given 2 dimensional shape in an optimized manner.

An Attract-Repel Decomposition of Undirected Networks arxiv:2106.09671 📈 2

Alexander Peysakhovich, Leon Bottou

**Abstract:** Dot product latent space embedding is a common form of representation learning in undirected graphs (e.g. social networks, co-occurrence networks). We show that such models have problems dealing with 'intransitive' situations where A is linked to B, B is linked to C but A is not linked to C. Such situations occur in social networks when opposites attract (heterophily) and in co-occurrence networks when there are substitute nodes (e.g. the presence of Pepsi or Coke, but rarely both, in otherwise similar purchase baskets). We present a simple expansion which we call the attract-repel (AR) decomposition: a set of latent attributes on which similar nodes attract and another set of latent attributes on which similar nodes repel. We demonstrate the AR decomposition in real social networks and show that it can be used to measure the amount of latent homophily and heterophily. In addition, it can be applied to co-occurrence networks to discover roles in teams and find substitutable ingredients in recipes.

Prototypical Graph Contrastive Learning arxiv:2106.09645 📈 2

Shuai Lin, Pan Zhou, Zi-Yuan Hu, Shuojia Wang, Ruihui Zhao, Yefeng Zheng, Liang Lin, Eric Xing, Xiaodan Liang

**Abstract:** Graph-level representations are critical in various real-world applications, such as predicting the properties of molecules. But in practice, precise graph annotations are generally very expensive and time-consuming. To address this issue, graph contrastive learning constructs instance discrimination task which pulls together positive pairs (augmentation pairs of the same graph) and pushes away negative pairs (augmentation pairs of different graphs) for unsupervised representation learning. However, since for a query, its negatives are uniformly sampled from all graphs, existing methods suffer from the critical sampling bias issue, i.e., the negatives likely having the same semantic structure with the query, leading to performance degradation. To mitigate this sampling bias issue, in this paper, we propose a Prototypical Graph Contrastive Learning (PGCL) approach. Specifically, PGCL models the underlying semantic structure of the graph data via clustering semantically similar graphs into the same group, and simultaneously encourages the clustering consistency for different augmentations of the same graph. Then given a query, it performs negative sampling via drawing the graphs from those clusters that differ from the cluster of query, which ensures the semantic difference between query and its negative samples. Moreover, for a query, PGCL further reweights its negative samples based on the distance between their prototypes (cluster centroids) and the query prototype such that those negatives having moderate prototype distance enjoy relatively large weights. This reweighting strategy is proved to be more effective than uniform sampling. Experimental results on various graph benchmarks testify the advantages of our PGCL over state-of-the-art methods.

Meta-Calibration: Meta-Learning of Model Calibration Using Differentiable Expected Calibration Error arxiv:2106.09613 📈 2

Ondrej Bohdal, Yongxin Yang, Timothy Hospedales

**Abstract:** Calibration of neural networks is a topical problem that is becoming increasingly important for real-world use of neural networks. The problem is especially noticeable when using modern neural networks, for which there is significant difference between the model confidence and the confidence it should have. Various strategies have been successfully proposed, yet there is more space for improvements. We propose a novel approach that introduces a differentiable metric for expected calibration error and successfully uses it as an objective for meta-learning, achieving competitive results with state-of-the-art approaches. Our approach presents a new direction of using meta-learning to directly optimize model calibration, which we believe will inspire further work in this promising and new direction.

Controllable Confidence-Based Image Denoising arxiv:2106.09311 📈 2

Haley Owsianko, Florian Cassayre, Qiyuan Liang

**Abstract:** Image denoising is a classic restoration problem. Yet, current deep learning methods are subject to the problems of generalization and interpretability. To mitigate these problems, in this project, we present a framework that is capable of controllable, confidence-based noise removal. The framework is based on the fusion between two different denoised images, both derived from the same noisy input. One of the two is denoised using generic algorithms (e.g. Gaussian), which make few assumptions on the input images, therefore, generalize in all scenarios. The other is denoised using deep learning, performing well on seen datasets. We introduce a set of techniques to fuse the two components smoothly in the frequency domain. Beyond that, we estimate the confidence of a deep learning denoiser to allow users to interpret the output, and provide a fusion strategy that safeguards them against out-of-distribution inputs. Through experiments, we demonstrate the effectiveness of the proposed framework in different use cases.

Cooperative Multi-Agent Reinforcement Learning Based Distributed Dynamic Spectrum Access in Cognitive Radio Networks arxiv:2106.09274 📈 2

Xiang Tan, Li Zhou, Haijun Wang, Yuli Sun, Haitao Zhao, Boon-Chong Seet, Jibo Wei, Victor C. M. Leung

**Abstract:** With the development of the 5G and Internet of Things, amounts of wireless devices need to share the limited spectrum resources. Dynamic spectrum access (DSA) is a promising paradigm to remedy the problem of inefficient spectrum utilization brought upon by the historical command-and-control approach to spectrum allocation. In this paper, we investigate the distributed DSA problem for multi-user in a typical multi-channel cognitive radio network. The problem is formulated as a decentralized partially observable Markov decision process (Dec-POMDP), and we proposed a centralized off-line training and distributed on-line execution framework based on cooperative multi-agent reinforcement learning (MARL). We employ the deep recurrent Q-network (DRQN) to address the partial observability of the state for each cognitive user. The ultimate goal is to learn a cooperative strategy which maximizes the sum throughput of cognitive radio network in distributed fashion without coordination information exchange between cognitive users. Finally, we validate the proposed algorithm in various settings through extensive experiments. From the simulation results, we can observe that the proposed algorithm can converge fast and achieve almost the optimal performance.

Evo* 2021 -- Late-Breaking Abstracts Volume arxiv:2106.11804 📈 1

A. M. Mora, A. I. Esparcia-Alcázar

**Abstract:** Volume with the Late-Breaking Abstracts submitted to the Evo* 2021 Conference, held online from 7 to 9 of April 2021. These papers present ongoing research and preliminary results investigating on the application of different approaches of Bioinspired Methods (mainly Evolutionary Computation) to different problems, most of them real world ones.

Sphynx: ReLU-Efficient Network Design for Private Inference arxiv:2106.11755 📈 1

Minsu Cho, Zahra Ghodsi, Brandon Reagen, Siddharth Garg, Chinmay Hegde

**Abstract:** The emergence of deep learning has been accompanied by privacy concerns surrounding users' data and service providers' models. We focus on private inference (PI), where the goal is to perform inference on a user's data sample using a service provider's model. Existing PI methods for deep networks enable cryptographically secure inference with little drop in functionality; however, they incur severe latency costs, primarily caused by non-linear network operations (such as ReLUs). This paper presents Sphynx, a ReLU-efficient network design method based on micro-search strategies for convolutional cell design. Sphynx achieves Pareto dominance over all existing private inference methods on CIFAR-100. We also design large-scale networks that support cryptographically private inference on Tiny-ImageNet and ImageNet.

$α$-Stable convergence of heavy-tailed infinitely-wide neural networks arxiv:2106.11064 📈 1

Paul Jung, Hoil Lee, Jiho Lee, Hongseok Yang

**Abstract:** We consider infinitely-wide multi-layer perceptrons (MLPs) which are limits of standard deep feed-forward neural networks. We assume that, for each layer, the weights of an MLP are initialized with i.i.d. samples from either a light-tailed (finite variance) or heavy-tailed distribution in the domain of attraction of a symmetric $α$-stable distribution, where $α\in(0,2]$ may depend on the layer. For the bias terms of the layer, we assume i.i.d. initializations with a symmetric $α$-stable distribution having the same $α$ parameter of that layer. We then extend a recent result of Favaro, Fortini, and Peluchetti (2020), to show that the vector of pre-activation values at all nodes of a given hidden layer converges in the limit, under a suitable scaling, to a vector of i.i.d. random variables with symmetric $α$-stable distributions.

Topological Indoor Mapping through WiFi Signals arxiv:2106.09789 📈 1

Bastian Schaefermeier, Gerd Stumme, Tom Hanika

**Abstract:** The ubiquitous presence of WiFi access points and mobile devices capable of measuring WiFi signal strengths allow for real-world applications in indoor localization and mapping. In particular, no additional infrastructure is required. Previous approaches in this field were, however, often hindered by problems such as effortful map-building processes, changing environments and hardware differences. We tackle these problems focussing on topological maps. These represent discrete locations, such as rooms, and their relations, e.g., distances and transition frequencies. In our unsupervised method, we employ WiFi signal strength distributions, dimension reduction and clustering. It can be used in settings where users carry mobile devices and follow their normal routine. We aim for applications in short-lived indoor events such as conferences.

Gradient-free optimization of chaotic acoustics with reservoir computing arxiv:2106.09780 📈 1

Francisco Huhn, Luca Magri

**Abstract:** We develop a versatile optimization method, which finds the design parameters that minimize time-averaged acoustic cost functionals. The method is gradient-free, model-informed, and data-driven with reservoir computing based on echo state networks. First, we analyse the predictive capabilities of echo state networks both in the short- and long-time prediction of the dynamics. We find that both fully data-driven and model-informed architectures learn the chaotic acoustic dynamics, both time-accurately and statistically. Informing the training with a physical reduced-order model with one acoustic mode markedly improves the accuracy and robustness of the echo state networks, whilst keeping the computational cost low. Echo state networks offer accurate predictions of the long-time dynamics, which would be otherwise expensive by integrating the governing equations to evaluate the time-averaged quantity to optimize. Second, we couple echo state networks with a Bayesian technique to explore the design thermoacoustic parameter space. The computational method is minimally intrusive. Third, we find the set of flame parameters that minimize the time-averaged acoustic energy of chaotic oscillations, which are caused by the positive feedback with a heat source, such as a flame in gas turbines or rocket motors. These oscillations are known as thermoacoustic oscillations. The optimal set of flame parameters is found with the same accuracy as brute-force grid search, but with a convergence rate that is more than one order of magnitude faster. This work opens up new possibilities for non-intrusive (``hands-off'') optimization of chaotic systems, in which the cost of generating data, for example from high-fidelity simulations and experiments, is high.

A deep generative model for probabilistic energy forecasting in power systems: normalizing flows arxiv:2106.09370 📈 1

Jonathan Dumas, Antoine Wehenkel Damien Lanaspeze, Bertrand Cornélusse, Antonio Sutera

**Abstract:** Greater direct electrification of end-use sectors with a higher share of renewables is one of the pillars to power a carbon-neutral society by 2050. However, in contrast to conventional power plants, renewable energy is subject to uncertainty raising challenges for their interaction with power systems. Scenario-based probabilistic forecasting models have become a vital tool to equip decision-makers. This paper presents to the power systems forecasting practitioners a recent deep learning technique, the normalizing flows, to produce accurate scenario-based probabilistic forecasts that are crucial to face the new challenges in power systems applications. The strength of this technique is to directly learn the stochastic multivariate distribution of the underlying process by maximizing the likelihood. Through comprehensive empirical evaluations using the open data of the Global Energy Forecasting Competition 2014, we demonstrate that this methodology is competitive with other state-of-the-art deep learning generative models: generative adversarial networks and variational autoencoders. The models producing weather-based wind, solar power, and load scenarios are properly compared in terms of forecast value by considering the case study of an energy retailer and quality using several complementary metrics. The numerical experiments are simple and easily reproducible. Thus, we hope it will encourage other forecasting practitioners to test and use normalizing flows in power system applications such as bidding on electricity markets, scheduling power systems with high renewable energy sources penetration, energy management of virtual power plan or microgrids, and unit commitment.

PEN4Rec: Preference Evolution Networks for Session-based Recommendation arxiv:2106.09306 📈 1

Dou Hu, Lingwei Wei, Wei Zhou, Xiaoyong Huai, Zhiqi Fang, Songlin Hu

**Abstract:** Session-based recommendation aims to predict user the next action based on historical behaviors in an anonymous session. For better recommendations, it is vital to capture user preferences as well as their dynamics. Besides, user preferences evolve over time dynamically and each preference has its own evolving track. However, most previous works neglect the evolving trend of preferences and can be easily disturbed by the effect of preference drifting. In this paper, we propose a novel Preference Evolution Networks for session-based Recommendation (PEN4Rec) to model preference evolving process by a two-stage retrieval from historical contexts. Specifically, the first-stage process integrates relevant behaviors according to recent items. Then, the second-stage process models the preference evolving trajectory over time dynamically and infer rich preferences. The process can strengthen the effect of relevant sequential behaviors during the preference evolution and weaken the disturbance from preference drifting. Extensive experiments on three public datasets demonstrate the effectiveness and superiority of the proposed model.

Brushless Motor Performance Optimization by Eagle Strategy with Firefly and PSO arxiv:2106.11135 📈 0

Appalabathula Venkatesh, Pradeepa H, Chidanandappa R, Shankar Nalinakshan, Jayasankar V N

**Abstract:** Brushless motors has special place though different motors are available because of its special features like absence in commutation, reduced noise and longer lifetime etc., The experimental parameter tracking of BLDC Motor can be achieved by developing a Reference system and their stability is guaranteed by adopting Lyapunov Stability theorems. But the stability is guaranteed only if the adaptive system is incorporated with the powerful and efficient optimization techniques. In this paper the powerful eagle strategy with Particle Swarm optimization and Firefly algorithms are applied to evaluate the performance of brushless motor Where, Eagle Strategy(ES) with the use of Levys walk distribution function performs diversified global search and the Particle Swarm Optimization (PSO) and Firefly Algorithm(FFA) performs the efficient intensive local search. The combined operation makes the overall optimization technique as much convenient The simulation results are obtained by using MATLAB Simulink software

Disinformation, Stochastic Harm, and Costly Effort: A Principal-Agent Analysis of Regulating Social Media Platforms arxiv:2106.09847 📈 0

Shehroze Khan, James R. Wright

**Abstract:** The spread of disinformation on social media platforms is harmful to society. This harm may manifest as a gradual degradation of public discourse; but it can also take the form of sudden dramatic events such as the recent insurrection on Capitol Hill. The platforms themselves are in the best position to prevent the spread of disinformation, as they have the best access to relevant data and the expertise to use it. However, mitigating disinformation is costly, not only for implementing detection algorithms or employing manual effort, but also because limiting such highly viral content impacts user engagement and thus potential advertising revenue. Since the costs of harmful content are borne by other entities, the platform will therefore have no incentive to exercise the socially-optimal level of effort. This problem is similar to that of environmental regulation, in which the costs of adverse events are not directly borne by a firm, the mitigation effort of a firm is not observable, and the causal link between a harmful consequence and a specific failure is difficult to prove. For environmental regulation, one solution is to perform costly monitoring to ensure that the firm takes adequate precautions according to a specified rule. However, a fixed rule for classifying disinformation becomes less effective over time, as bad actors can learn to sequentially and strategically bypass it. Encoding our domain as a Markov decision process, we demonstrate that no penalty based on a static rule, no matter how large, can incentivize adequate effort. Penalties based on an adaptive rule can incentivize optimal effort, but counterintuitively, only if the regulator sufficiently overreacts to harmful events by requiring a greater-than-optimal level of effort. We prescribe the design of mechanisms that elicit platforms' costs of precautionary effort relating to the control of disinformation.

Private Federated Learning Without a Trusted Server: Optimal Algorithms for Convex Losses arxiv:2106.09779 📈 0

Andrew Lowy, Meisam Razaviyayn

**Abstract:** This paper studies the problem of federated learning (FL) in the absence of a trustworthy server/clients. In this setting, each client needs to ensure the privacy of its own data without relying on the server or other clients. We study local differential privacy (LDP) at the client level and provide tight upper and lower bounds that establish the minimax optimal rates (up to logarithms) for LDP convex/strongly convex federated stochastic optimization. Our rates match the optimal statistical rates in certain practical parameter regimes ("privacy for free"). Second, we develop an accelerated distributed noisy SGD algorithm, leading to the first non-trivial LDP risk bounds for FL with non-i.i.d. clients. Third, we consider the special case where each client's loss function is empirical and use a variation of our accelerated LDP FL algorithm to improve communication complexity compared to existing works. We also provide matching lower bounds, establishing the optimality of our algorithm for convex/strongly convex settings. Fourth, with a secure shuffler to anonymize client reports (but without a trusted server), our algorithm attains the optimal central DP rates for stochastic convex/strongly convex optimization, thereby achieving optimality in the local and central models simultaneously. Our upper bounds quantify the role of network communication reliability in performance.

CIRA Guide to Custom Loss Functions for Neural Networks in Environmental Sciences -- Version 1 arxiv:2106.09757 📈 0

Imme Ebert-Uphoff, Ryan Lagerquist, Kyle Hilburn, Yoonjin Lee, Katherine Haynes, Jason Stock, Christina Kumler, Jebb Q. Stewart

**Abstract:** Neural networks are increasingly used in environmental science applications. Furthermore, neural network models are trained by minimizing a loss function, and it is crucial to choose the loss function very carefully for environmental science applications, as it determines what exactly is being optimized. Standard loss functions do not cover all the needs of the environmental sciences, which makes it important for scientists to be able to develop their own custom loss functions so that they can implement many of the classic performance measures already developed in environmental science, including measures developed for spatial model verification. However, there are very few resources available that cover the basics of custom loss function development comprehensively, and to the best of our knowledge none that focus on the needs of environmental scientists. This document seeks to fill this gap by providing a guide on how to write custom loss functions targeted toward environmental science applications. Topics include the basics of writing custom loss functions, common pitfalls, functions to use in loss functions, examples such as fractions skill score as loss function, how to incorporate physical constraints, discrete and soft discretization, and concepts such as focal, robust, and adaptive loss. While examples are currently provided in this guide for Python with Keras and the TensorFlow backend, the basic concepts also apply to other environments, such as Python with PyTorch. Similarly, while the sample loss functions provided here are from meteorology, these are just examples of how to create custom loss functions. Other fields in the environmental sciences have very similar needs for custom loss functions, e.g., for evaluating spatial forecasts effectively, and the concepts discussed here can be applied there as well. All code samples are provided in a GitHub repository.

On Anytime Learning at Macroscale arxiv:2106.09563 📈 0

Lucas Caccia, Jing Xu, Myle Ott, Marc'Aurelio Ranzato, Ludovic Denoyer

**Abstract:** Classical machine learning frameworks assume access to a possibly large dataset in order to train a predictive model. In many practical applications however, data does not arrive all at once, but in batches over time. This creates a natural trade-off between accuracy of a model and time to obtain such a model. A greedy predictor could produce non-trivial predictions by immediately training on batches as soon as these become available but, it may also make suboptimal use of future data. On the other hand, a tardy predictor could wait for a long time to aggregate several batches into a larger dataset, but ultimately deliver a much better performance. In this work, we consider such a streaming learning setting, which we dub anytime learning at macroscale} (ALMA). It is an instance of anytime learning applied not at the level of a single chunk of data, but at the level of the entire sequence of large batches. We first formalize this learning setting, we then introduce metrics to assess how well learners perform on the given task for a given memory and compute budget, and finally we test about thirty baseline approaches on three standard benchmarks repurposed for anytime learning at macroscale. Our findings indicate that no model strikes the best trade-off across the board. While replay-based methods attain the lowest error rate, they also incur in a 5 to 10 times increase of compute. Approaches that grow capacity over time do offer better scaling in terms of training flops, but they also underperform simpler ensembling methods in terms of error rate. Overall, ALMA offers both a good abstraction of the typical learning setting faced everyday by practitioners, and a set of unsolved modeling problems for those interested in efficient learning of dynamic models.

On Deep Neural Network Calibration by Regularization and its Impact on Refinement arxiv:2106.09385 📈 0

Aditya Singh, Alessandro Bay, Biswa Sengupta, Andrea Mirabile

**Abstract:** Deep neural networks have been shown to be highly miscalibrated. often they tend to be overconfident in their predictions. It poses a significant challenge for safety-critical systems to utilise deep neural networks (DNNs), reliably. Many recently proposed approaches to mitigate this have demonstrated substantial progress in improving DNN calibration. However, they hardly touch upon refinement, which historically has been an essential aspect of calibration. Refinement indicates separability of a network's correct and incorrect predictions. This paper presents a theoretically and empirically supported exposition reviewing refinement of a calibrated model. Firstly, we show the breakdown of expected calibration error (ECE), into predicted confidence and refinement under the assumption of over-confident predictions. Secondly, linking with this result, we highlight that regularization based calibration only focuses on naively reducing a model's confidence. This logically has a severe downside to a model's refinement as correct and incorrect predictions become tightly coupled. Lastly, connecting refinement with ECE also provides support to existing refinement based approaches which improve calibration but do not explain the reasoning behind it. We support our claims through rigorous empirical evaluations of many state of the art calibration approaches on widely used datasets and neural networks. We find that many calibration approaches with the likes of label smoothing, mixup etc. lower the usefulness of a DNN by degrading its refinement. Even under natural data shift, this calibration-refinement trade-off holds for the majority of calibration methods.

Federated Learning Framework with Straggling Mitigation and Privacy-Awareness for AI-based Mobile Application Services arxiv:2106.09261 📈 0

Yuris Mulya Saputra, Diep N. Nguyen, Dinh Thai Hoang, Quoc-Viet Pham, Eryk Dutkiewicz, Won-Joo Hwang

**Abstract:** In this work, we propose a novel framework to address straggling and privacy issues for federated learning (FL)-based mobile application services, taking into account limited computing/communications resources at mobile users (MUs)/mobile application provider (MAP), privacy cost, the rationality and incentive competition among MUs in contributing data to the MAP. Particularly, the MAP first determines a set of the best MUs for the FL process based on the MUs' provided information/features. To mitigate straggling problems with privacy-awareness, each selected MU can then encrypt part of local data and upload the encrypted data to the MAP for an encrypted training process, in addition to the local training process. For that, each selected MU can propose a contract to the MAP according to its expected trainable local data and privacy-protected encrypted data. To find the optimal contracts that can maximize utilities of the MAP and all the participating MUs while maintaining high learning quality of the whole system, we first develop a multi-principal one-agent contract-based problem leveraging FL-based multiple utility functions. These utility functions account for the MUs' privacy cost, the MAP's limited computing resources, and asymmetric information between the MAP and MUs. Then, we transform the problem into an equivalent low-complexity problem and develop a light-weight iterative algorithm to effectively find the optimal solutions. Experiments with a real-world dataset show that our framework can speed up training time up to 49% and improve prediction accuracy up to 4.6 times while enhancing the network's social welfare, i.e., total utility of all participating entities, up to 114% under the privacy cost consideration compared with those of baseline methods.

Next Page