paper_name stringlengths 11 170 | text stringlengths 8.07k 307k | summary stringlengths 152 6.16k | paper_id stringlengths 43 43 |
|---|---|---|---|
Representation Balancing Offline Model-based Reinforcement Learning | 1 INTRODUCTION . Reinforcement learning ( RL ) has accomplished remarkable results in a wide range of domains , but its successes were mostly based on a large number of online interactions with the environment . However , in many real-world tasks , exploratory online interactions are either very expensive or dangerous ... | In this paper, the authors propose a model-based approach with representation balancing (RepB-SDE)to cope with the distribution shift of offline reinforcement learning. RepB-SDE learns a robust representation for the model learning process, which regularizes the distance between the data distribution and the discount s... | SP:561e202e2167aa602c815441e8ca52992d81b03b |
Information distance for neural network functions | 1 INTRODUCTION . Deep neural networks can be trained to represent complex functions that describe sophisticated input-output relationships , such as image classification and machine translation . Because the functions are highly non-linear and are parameterized in high-dimensional spaces , there is relatively little un... | This paper explores the problem of designing a distance measure in the space of the functions parameterized by neural networks. Ideally, such a measure should be independent of the parameterization of the networks. Also, the measure should support quantifying the distance between the networks with different structures ... | SP:7689dac5ea0db9c2a021e33f03d8cdeb9b5e4290 |
Exploiting Verified Neural Networks via Floating Point Numerical Error | 1 INTRODUCTION . Deep neural networks ( DNNs ) are known to be vulnerable to adversarial inputs ( Szegedy et al. , 2014 ) , which are images , audio , or texts indistinguishable to human perception that cause a DNN to give substantially different results . This situation has motivated the development of network verific... | In the recent literature there has been a rise in the number of papers which attempt to verify neural networks. The specification of the verification problems often gets adapted according to the application in mind. More specifically, for image classification networks, the problem is to prove that the output of the neu... | SP:3f8deffff011d2fb7cc8d38f8e7e28ede4e632ca |
Exchanging Lessons Between Algorithmic Fairness and Domain Generalization | 1 INTRODUCTION . Machine learning achieves super-human performance on many tasks when the test data is drawn from the same distribution as the training data . However , when the two distributions differ , model performance can severely degrade to even below chance predictions ( Geirhos et al. , 2020 ) . Tiny perturbati... | The main contribution of the paper is to highlight the similarity between two active areas in ML namely "domain generalization" and "fairness". Further, the paper proposes an approach inspired by recent developments in the fairness literature for domain generalization. The high-level idea is that similarly to the way t... | SP:efea29871d33fd89de348bc243a5ee0265b2e051 |
Learning Deep Latent Variable Models via Amortized Langevin Dynamics | 1 INTRODUCTION . Latent variable models are widely used for generative modeling ( Bishop , 1998 ; Kingma & Welling , 2013 ) , principal component analysis ( Wold et al. , 1987 ) , and factor analysis ( Harman , 1976 ) . To learn a latent variable model , it is essential to estimate the latent variables , z , from the o... | In this paper, the author presented an advanced autoencoder framework LAE. Instead of element-wise MCMC, LAE collected samples from the posterior using the amortized Langevin dynamics of a potential energy distribution. In CLAE, an extended version of LAE, the author used an intractable energy function as the prior, an... | SP:ca9a9e8d0066ca55d4cd760df661bec09cdeb8eb |
The Intrinsic Dimension of Images and Its Impact on Learning | 1 INTRODUCTION . The idea that real-world data distributions can be described by very few variables underpins machine learning research from manifold learning to dimension reduction ( Besold & Spokoiny , 2019 ; Fodor , 2002 ) . The number of variables needed to describe a data distribution is known as its intrinsic dim... | The authors report a novel application of GANs to validate the maximum likelihood estimator (MLE) of the intrinsic dimension (ID) of image data sets. Then they use the MLE ID estimator to characterize the intrinsic dimension of several commonly used computer vision data sets, and link the data set ID to the generalizab... | SP:bf07fc882fe3aca4c50e07df79f22d4b8b3abb56 |
Improving Generalizability of Protein Sequence Models via Data Augmentations | While protein sequence data is an emerging application domain for machine learning methods , small modifications to protein sequences can result in difficult-topredict changes to the protein ’ s function . Consequently , protein machine learning models typically do not use randomized data augmentation procedures analog... | The paper explores the impact of different types of data augmentations for protein sequence data, and does a thorough benchmark analysis on them. The authors used a pre-trained transformer model, fine tuned the model on augmented data using two approaches, namely, contrastive learning and masked token prediction. This ... | SP:f1af5160de3da8d992ac6bba8fbb7b0086efdb12 |
MELR: Meta-Learning via Modeling Episode-Level Relationships for Few-Shot Learning | Most recent few-shot learning ( FSL ) approaches are based on episodic training whereby each episode samples few training instances ( shots ) per class to imitate the test condition . However , this strict adhering to test condition has a negative side effect , that is , the trained model is susceptible to the poor sam... | This paper proposes a way to exploit relationships across tasks in episodic training with the goal of improving the trained models who might be susceptible to poor sampling in for few-shot learning scenarios. The proposed model consists of two components: a cross-attention transformer (CEAM) which is used to observe de... | SP:52b51e46d40e554920d48625707a433db2dc233c |
TextTN: Probabilistic Encoding of Language on Tensor Network | 1 INTRODUCTION . Machine learning incorporating with the quantum mechanics forms a novel interdisciplinary field known as quantum machine learning ( Huggins et al. , 2019 ; Ran et al. , 2020 ) . Tensor network ( TN ) as a novel model has become prominent in the field of quantum machine learning ( Biamonte et al. , 2017... | The paper proposes a tensor network for text classification. There are two components: (i) word-GTNs convert word embeddings to m-d probability encoding vectors, and (ii) a sentence-DND takes the word probability encoding vectors as input, combining them using matrix product state (MPS). Experiments on several text cla... | SP:5ee24df635a6659978378a8ff6e0cc41e51b6010 |
HyperSAGE: Generalizing Inductive Representation Learning on Hypergraphs | 1 INTRODUCTION . Graphs are considered the most prevalent structures for discovering useful information within a network , especially because of their capability to combine object-level information with the underlying inter-object relations ( Wu et al. , 2020 ) . However , most structures encountered in practical appli... | In this paper, the authors study the problem of learning node embeddings for hypergraphs. While most of the existing studies consider reducing hyper-graphs into graphs, this paper studies learning embeddings directly on the hypergraphs using two stages of aggregations / sampling. The efficacy of the proposed method is ... | SP:6e300c6a4bdcf32b80cecf2d4526b99deb30912a |
Provable Memorization via Deep Neural Networks using Sub-linear Parameters | 1 INTRODUCTION . The modern trend of over-parameterizing neural networks has shifted the focus of deep learning theory from analyzing their expressive power toward understanding the generalization capabilities of neural networks . While the celebrated universal approximation theorems state that over-parameterization en... | The paper studies the memorization capacity of deep networks as a function of the number of parameters. Many prior works have shown that to memorize $N$ examples $O(N)$ parameters are sufficient and that to memorize any set of $N$ examples $\Omega(N)$ parameters are necessary. This work shows that under very mild and c... | SP:1d63d99b43018556784ecc4a3ee494c71e7ef06e |
Learning Reasoning Paths over Semantic Graphs for Video-grounded Dialogues | 1 INTRODUCTION . Traditional visual question answering ( Antol et al. , 2015 ; Jang et al. , 2017 ) involves answering questions about a given image . Extending from this line of research , recently Das et al . ( 2017 ) ; Alamri et al . ( 2019 ) add another level of complexity by positioning each question and answer pa... | This paper addresses the visual question answering in a multi-turn or conversational setting. Given a video (series of frames or images), a model has to reason across space and time to arrive at a correct answer for a given question. This task involves understanding the content and context of dialogue turns, i.e., give... | SP:aed4e9af07b32dc6f38e851db17287e7a29f6f09 |
Generalized Energy Based Models | 1 INTRODUCTION . Energy-based models ( EBMs ) have a long history in physics , statistics and machine learning ( LeCun et al. , 2006 ) . They belong to the class of explicit models , and can be described by a family of energies E which define probability distributions with density proportional to exp ( −E ) . Those mod... | This paper proposes a framework called GEBM that combine an implicit generator and an EBM to define a probabilistic model on low dimensional manifold. Specifically, the implicit generator defines the base distribution, and the EBM refines the base. Finally, this method is equivalent to define an EBM on the latent space... | SP:d8c1eee3aad4cbe04e5602c4a4da1da44a8ca9d3 |
Clairvoyance: A Pipeline Toolkit for Medical Time Series | Python Software Repository : https : //github.com/vanderschaarlab/clairvoyance 1 INTRODUCTION . Inference over time series is ubiquitous in medical problems [ 1–7 ] . With the increasing availability and accessibility of electronic patient records , machine learning for clinical decision support has made great strides ... | The authors present a new package aimed at improving the design and validation of pipelines using medical time series data. The pipeline covers many aspects of time series pipelines including pre-processing, prediction, treatment effect estimation, calibration, etc. The package, as depicted in the paper, appears to be ... | SP:3cc3f1b3e24923e2e84d0b9761e5fb30fa88bbeb |
PURE: An Uncertainty-aware Recommendation Framework for Maximizing Expected Posterior Utility of Platform | 1 INTRODUCTION . Commercial recommendation systems have been widely applied among prevalent content distribution platforms such as YouTube , TikTok , Amazon and Taobao . During the interactive process on the recommendation platform , the users may find contents of their interests and avoid the information overload prob... | The paper aims to model the posterior utility of showing an item given a static display policy, where the utility function captures both: (a) uncertainty over item dimensions from the user perspective, and (b) influence of the policy on the user. It is motivated by the fact that most recommender systems don't take into... | SP:415de370d5dea4aa3136d79bef9bf04f733d8285 |
Systematic Evaluation of Causal Discovery in Visual Model Based Reinforcement Learning | 1 INTRODUCTION . Deep learning methods have made immense progress on many reinforcement learning ( RL ) tasks in recent years . However , the performance of these methods still pales in comparison to human abilities in many cases . Contemporary deep reinforcement learning models have a ways to go to achieve robust gene... | This paper is a review of model-based approaches of integrating causal inference to reinforcement learning (RL) in different environments (application areas). The authors provide software to analyse how three types of models (“monolithic”, i.e. latent space models without a graph-like structure of the latent space, gra... | SP:86435186f0d117c14bbf6d300053dd46884ea061 |
N-Bref : A High-fidelity Decompiler Exploiting Programming Structures | 1 INTRODUCTION . Decompilation , which is a process of recovering source code from binary , is useful in many situations where it is necessary to analyze or understand software for which source code is not available . For example , decompilation is highly valuable in many security and forensics applications ( Lin et al... | The authors present a neural-based decompilation framework. They generate synthetic input data in order to make sure source code examples have a consistent code style that can be more easily learned. They predict types of variables and the actual code in two separate steps. For both steps, they employ a custom transfor... | SP:ec1ff351fa8fb2cd61f9662a9d0e7db6531fcb4f |
Meta Back-Translation | 1 INTRODUCTION . While Neural Machine Translation ( NMT ) delivers state-of-the-art performance across many translation tasks , this performance is usually contingent on the existence of large amounts of training data ( Sutskever et al. , 2014 ; Vaswani et al. , 2017 ) . Since large parallel training datasets are often... | This paper applies techniques from meta-learning to derive and end-to-end update rule for a workflow involving backtranslation, specificically maximizing translation performance of the forward model, while updating the backward model to produce backtranslations that are maximally useful to improve the forward model's q... | SP:0586aff632d77cbf60cefae509a93bc22c95655e |
Response Modeling of Hyper-Parameters for Deep Convolutional Neural Networks | 1 INTRODUCTION . The choice of Hyper-Parameters ( HP ) – such as initial learning rate , batch size , and weight decay – has shown to greatly impact the generalization performance of Deep Neural Network ( DNN ) training ( Keskar et al. , 2017 ; Wilson et al. , 2017 ; Li et al. , 2019 ; Yu & Zhu , 2020 ) . By increasing... | The paper proposes an efficient framework to search for the optimal initial learning rate to train neural networks. The key idea is to introduce Knowledge Gain, a metric derived from the singular values of each layer, to indicate the convergency quality of training. Taking advantage of the metric, a logarithmic grid se... | SP:ea8234f4533090e0cfe197ddb70f375f3ed49418 |
Training Federated GANs with Theoretical Guarantees: A Universal Aggregation Approach | 1 INTRODUCTION . Generative Adversarial Networks ( GANs ) have attracted much attention due to their ability to generate realistic-looking synthetic data ( Goodfellow et al. , 2014 ; Zhang et al. , 2018 ; Liu et al. , 2019b ; Shaham et al. , 2019 ; Dai et al. , 2017 ; Kumar et al. , 2017 ) . In order to obtain a powerf... | The paper proposes a new method, UA-GAN to train GANs in a federated learning setup. The method simulates a central discriminator D_ua such that the odds values of the central discriminator is equivalent to the weighted sum of the local discriminators. The central generator is then trained based on the simulated centra... | SP:3d65be849f99ab19e16eab84ba7cd7748d3ed8ad |
Parameter-Efficient Transfer Learning with Diff Pruning | 1 INTRODUCTION . Task-specific finetuning of pretrained deep networks has become the dominant paradigm in contemporary NLP , achieving state-of-the-art results across a suite of natural language understanding tasks ( Devlin et al. , 2019 ; Liu et al. , 2019c ; Yang et al. , 2019 ; Lan et al. , 2020 ) . While straightfo... | This paper proposes diff pruning, an alternative paradigm for parameter-efficient transfer learning of pre-trained models. Similar to adapters, diff pruning leaves the body of the pre-trained model unchanged. Rather than inserting additional task-specific parameters into the pre-trained model, diff pruning adds reparam... | SP:f08a22a7a003f9fff3d1366b923ed961489d9158 |
Interpretability Through Invertibility: A Deep Convolutional Network With Ideal Counterfactuals And Isosurfaces | Current state of the art computer vision applications rely on highly complex models . Their interpretability is mostly limited to post-hoc methods which are not guaranteed to be faithful to the model . To elucidate a model ’ s decision , we present a novel interpretable model based on an invertible deep convolutional n... | The paper presents a promising idea to build interpretable models by combining discriminative and generative approach. The proposed model uses an invertible neural network to model the data distribution. The invertibility helps in transforming the learned feature vector back to the image domain. A linear discriminative... | SP:83a6062cbcad4c8c40fe066abc7bd32a62f38b52 |
Impact-driven Exploration with Contrastive Unsupervised Representations | 1 INTRODUCTION . Reinforcement learning ( RL ) algorithms aim to learn an optimal policy that maximizes expected reward from the environment . The search for better RL algorithms is motivated by the fact that many complex real-world problems can be formulated as RL problems . Yet , environments with sparse rewards , wh... | The work provided a nice new method with some performance gains by combining several existing techniques. The presentation was clear and organized, with the new method getting both better performance and some improvements in interpretability. It provides a variety of visual analyses that are typical of this area of res... | SP:0b0a27b56520c182d6cdc92a338695f8a7813b83 |
Client Selection in Federated Learning: Convergence Analysis and Power-of-Choice Selection Strategies | 1 INTRODUCTION . Until recently , machine learning models were largely trained in the data center setting ( Dean et al. , 2012 ) using powerful computing nodes , fast inter-node communication links , and large centrally available training datasets . The future of machine learning lies in moving both data collection as ... | This work investigates federated optimization considering data heterogeneity, communication and computation limitations, and partial client participation. In contrast to past works, this paper focuses on deeper understanding of the effect of partial client participation on the convergence rate by considering biased cli... | SP:a25b3465107fa32de50e4457b87eba134792e07b |
Generalized Variational Continual Learning | 1 INTRODUCTION . Continual learning methods enable learning when a set of tasks changes over time . This topic is of practical interest as many real-world applications require models to be regularly updated as new data is collected or new tasks arise . Standard machine learning models and training procedures fail in th... | This paper proposes Generalized Variational Continual Learning (GVCL). It is shown that Online EWC and VCL are special cases of GVCL, along with other theoretical contributions. Further, GVCL is augmented with FiLM to alleviate weaknesses of VCL and GVCL. GVCL and GVCL-F are applied to a number of continual learning ta... | SP:40f435881d361a57f68c000a5cf06d868acbcda8 |
FedMix: Approximation of Mixup under Mean Augmented Federated Learning | Federated learning ( FL ) allows edge devices to collectively learn a model without directly sharing data within each device , thus preserving privacy and eliminating the need to store data globally . While there are promising results under the assumption of independent and identically distributed ( iid ) local data , ... | The paper proposed MAFL, a novel approach to conduct Mixup under the federated learning setting whiling preserving data privacy. The proposed FedMix scheme is inspired by Taylor’s expansion of the global Mixup formulation. The effectiveness of MAFL is justified via empirical studies over a simulated federated learning ... | SP:2d435fd5053bd60dd56049e2177e4cc8f5218759 |
Cortico-cerebellar networks as decoupled neural interfaces | 1 INTRODUCTION . Efficient credit assignment in the brain is a critical part of learning . However , how the brain solves the credit assignment problem remains a mystery . One of the central issues of credit assignment across multiple stages of processing is the need to wait for previous stages to finish their computat... | The authors proposed that cerebellum in the brain computes synthetic gradients, as used in decoupled neural interfaces (DNI), to enable learning in neural circuits without waiting for gradients to propagate backwards. The authors incorporated several architectural properties of biological cerebellum into their cerebell... | SP:332c59e494e36b043e48760cb2dfac206cafdcec |
Ruminating Word Representations with Random Noise Masking | 1 Introduction Most machine learning methodologies can be formulated to get computational representations from real-life objects ( e.g. , images , languages , and sounds ) and then get high-level representations using model architectures . Therefore , there have been two main approaches to improve model performances : ... | It's already known that embeddings like word2vec and glove are biased [1] and needs postprocessing for better performance. This paper designed a novel approach to do embedding normalizations. After each round of training, noise is intentionally introduced to perturbate the finetuned parameters. Afterwards, another roun... | SP:1292de91b0e7ab81457f925f72022d83ec061cc6 |
Improving the accuracy of neural networks in analog computing-in-memory systems by a generalized quantization method | 1 INTRODUCTION . Deep neural networks ( DNNs ) have been widely used in a variety of fields , such as computer vision ( Krizhevsky et al. , 2012 ; Simonyan & Zisserman , 2015 ; He et al. , 2016 ) , speech recognition ( Graves et al. , 2013 ; Hinton et al. , 2012 ; Graves & Jaitly , 2014 ) , natural language processing ... | This paper proposes a method to train weight-quantized neural networks. The authors propose to directly calculate the endpoints that minimize the quantization error according to the weight distribution of each layer. Empirical results on image classification tasks and object detection tasks show that the proposed meth... | SP:1d5d07627d5218eea719362a51fd1175bc2f841e |
Understanding the effects of data parallelism and sparsity on neural network training | 1 INTRODUCTION . Data parallelism is a straightforward and common approach to accelerate neural network training by processing training data in parallel using distributed systems . Being model-agnostic , it is applicable to training any neural networks , and the degree of parallelism equates to the size of mini-batch f... | The manuscript studies the effect of batch size at different sparsity levels (achieved by applying connection sensitivity pruning) on the required number of optimisation steps to reach a certain accuracy. The goal is to understand the interplay between those fundamental parameters. The empirical evaluation is performed... | SP:1a2c88b471d463d79a172c254483d8c92314fe3b |
Contrastive Code Representation Learning | 1 INTRODUCTION . Programmers increasingly rely on machine-aided programming tools to aid software development ( Kim et al. , 2012 ) . However , the wide diversity of programs encountered in practice limits the generalization of hand-written rules . Catching semantic bugs such as naming errors requires deeper language u... | This paper studies the self-supervised code functional representation learning and proposes a method called ContraCode. ContraCode utilizes some code functionality invariant transformations to generate positive pairs from the same code and negative pairs from different codes. After that, these codes pairs will be used ... | SP:e30f87da31dcb7e7ee9dd0abd503731d11d5160a |
Evaluating representations by the complexity of learning low-loss predictors | 1 INTRODUCTION . One of the first steps in building a machine learning system is selecting a representation of data . Whereas classical machine learning pipelines often begin with feature engineering , the advent of deep learning has led many to argue for pure end-to-end learning where the deep network constructs the f... | The submission addresses the problem of representation evaluation from the perspective of efficient learning of downstream predictors. Leveraging the introduced loss-data curve framework, the paper studies and demonstrates the limitations of the existing methods in terms of their implicit dependency on evaluation datas... | SP:5682a82e8671bdd5dee966273b981f63b4eebf2d |
An Algorithm for Out-Of-Distribution Attack to Neural Network Encoder | Deep neural networks ( DNNs ) , especially convolutional neural networks , have achieved superior performance on image classification tasks . However , such performance is only guaranteed if the input to a trained model is similar to the training samples , i.e. , the input follows the probability distribution of the tr... | The paper presents a method that attacks existing out-of-distribution (OOD) detection methods. Most of the existing OOD detection methods perform detection using a latent representation. Main motivation of the paper is that the size of the latent representation is much smaller than the input images which results mappin... | SP:193b21c862d83fd412bd5a07f49ca62e7285f62d |
Inferring Principal Components in the Simplex with Multinomial Variational Autoencoders | INFERRING PRINCIPAL COMPONENTS IN THE SIMPLEX WITH MULTINOMIAL VARIATIONAL AUTOENCODERS . Anonymous authors Paper under double-blind review 1 ABSTRACT . Covariance estimation on high-dimensional data is a central challenge across multiple scientific disciplines . Sparse high-dimensional count data , frequently encounte... | This paper extends prior results, namely that VAEs are able to learn the principal components. The novelty is the extension to a new distribution: multinomial logistic-normal distribution. This is achieved by using the Isometric log-ratio (ILR) transform. While prior results were derived analytically, this paper provid... | SP:19a28a50180cda10be3344064701fee76f354cf9 |
Adaptive Self-training for Neural Sequence Labeling with Few Labels | 1 INTRODUCTION . Motivation . Deep neural networks typically require large amounts of training data to achieve stateof-the-art performance . Recent advances with pre-trained language models like BERT ( Devlin et al. , 2019 ) , GPT-2 ( Radford et al. , 2019 ) and RoBERTa ( Liu et al. , 2019 ) have reduced this annotatio... | This paper proposes an adaptive self-training framework, called MetaST, for tackling few-shot sequence labeling tasks. The framework consists of several components: a teacher model that finetunes with the few-shot training data and generates noisy labels for the unlabeled examples; a student model that learns from re-w... | SP:dd0782278b556d2946ddd4bb7ea71c2bfbea948d |
GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding | Neural network scaling has been critical for improving the model quality in many real-world machine learning applications with vast amounts of training data and compute . Although this trend of scaling is affirmed to be a sure-fire approach for better model quality , there are challenges on the path such as the computa... | The paper applies mixture-of-experts (MoE) [1] to the Transformers to significantly increase the number of parameters in the model while keeping the total computational cost feasible. The main differences with [1] are: (1) Only choose 2 experts at each timestep; (2) Set a capacity upper bound for each expert to make su... | SP:f76f1289d7b47dd1bd381108f5b86a410613af9e |
Fine-Tuning Offline Reinforcement Learning with Model-Based Policy Optimization | 1 INTRODUCTION . Deep reinforcement learning has recently been able to achieve impressive results in a variety of video games ( Badia et al. , 2020 ) and board games ( Schrittwieser et al. , 2020 ) . However , it has had limited success in complicated real-world tasks . In contrast , deep supervised learning algorithms... | This paper proposes an improved offline RL (batch RL) algorithm combining the state-of-the-art behavior-regularization actor-critic method (Nair et al., 2020) with a model-based RL technique. N-trained probabilistic dynamics models generate fictitious trajectories with uncertainty-penalized rewards after pretraining th... | SP:36c53ab1d8f25c8c61d8b1538ed304b710c14849 |
Conditional Generative Modeling for De Novo Hierarchical Multi-Label Functional Protein Design | 1 INTRODUCTION . Designing proteins with a target biological function is an important task in biotechnology with highimpact implications in pharmaceutical research , such as in drug design or synthetic biology ( Huang et al. , 2016 ) . However , the task is challenging since the sequence-structure-function relationship... | In this manuscript, the authors present a conditional GAN for generating protein sequences given specified GO terms. They argue that this approach to conditional protein generation is more appropriate than sequence-based generation, because it gets directly at functional specification. At a high level, this is an inter... | SP:c5a5db22e2ac2eaa16a74238256753e567b07d9a |
AC-VAE: Learning Semantic Representation with VAE for Adaptive Clustering | 1 INTRODUCTION Unsupervised representation learning is a long-standing interest in the field of machine learning ( Peng et al. , 2016a ; Chen et al. , 2016 ; 2018 ; Deng et al. , 2019 ; Peng et al. , 2016b ) , which offers a promising way to scale-up the usable data amount for the current artificial intelligence method... | This paper proposes an adaptive neighbor clustering method by estimating normal distribution on the representation space. The proposed neighbor clustering can utilize the acceptable range for each dimension of each instance from the estimated variance, which leads to different size of neighbors for each instance, and i... | SP:9ebf89e9e24ce1a745f97b9d33bb5ec9979e60e5 |
Multi-View Disentangled Representation | 1 INTRODUCTION . Multi-view representation learning ( MRL ) involves learning representations by effectively leveraging information from different perspectives . The representations produced by MRL are effective when correlations across different views are accurately modeled and thus properly exploited for downstream t... | The goal of this paper is to define multi-view disentanglement in an unsupervised manner. The authors list four principal rules for representation disentanglement, including completeness of combining shared and specific representations, the exclusivity of two specific representations and between specific and shared rep... | SP:751b7dfd3e7de8e6f533137fc9ae7a65583c09e0 |
You Only Need Adversarial Supervision for Semantic Image Synthesis | ∗Equal contribution . Correspondence to { edgar.schoenfeld , vadim.sushko } @ bosch.com . 1 INTRODUCTION . Conditional generative adversarial networks ( GANs ) ( Mirza & Osindero , 2014 ) synthesize images conditioned on class labels ( Zhang et al. , 2019 ; Brock et al. , 2019 ) , text ( Reed et al. , 2016 ; Zhang et a... | In this paper, the authors approach the problem of conditional image generation via generative adversarial networks. To this end, they propose an approach that utilizes only semantic segmentation annotations and adversarial loss. No perceptual loss is required. Their discriminator leverages semantic labels to improve t... | SP:cef0728e41977750c56af5228b0e0dff4ec13358 |
Parametric Density Estimation with Uncertainty using Deep Ensembles | In parametric density estimation , the parameters of a known probability density1 are typically recovered from measurements by maximizing the log-likelihood.2 Prior knowledge of measurement uncertainties is not included in this method , po-3 tentially producing degraded or even biased parameter estimates . We propose4 ... | This paper addresses the problem of estimating the distribution parameters of features extracted from a set of high dimensional observations, a problem that is common in the physical sciences. To solve this problem, the authors present a deep learning approach that utilises a combination of (i) deep ensemble training, ... | SP:f4fafd66830ad4d90a13395ac5327de33d127a73 |
Energy-Based Models for Continual Learning | 1 INTRODUCTION . Humans are able to rapidly learn new skills and continuously integrate them with prior knowledge . The field of Continual Learning ( CL ) seeks to build artificial agents with the same capabilities ( Parisi et al. , 2019 ) . In recent years , CL has seen increased attention , particularly in the contex... | This paper explores the usage of EBMs in continual learning for classification. Although the application of EBMs in continual learning is novel, the general idea is a special case of the usage of EBMs for structured prediction, which has been widely studied. For instance, multi-class classification can be considered as... | SP:ff7570a39b118ef58a9bf05561824c85c5b48535 |
Variational Structured Attention Networks for Dense Pixel-Wise Prediction | 1 INTRODUCTION . Over the past decade , convolutional neural networks ( CNNs ) have become the privileged methodology to address computer vision tasks requiring dense pixel-wise prediction , such as semantic segmentation ( Chen et al. , 2016b ; Fu et al. , 2019 ) , monocular depth prediction ( Liu et al. , 2015 ; Roy &... | This paper proposes the VarIational STructured Attention networks (VISTA-Net), which improves pervious SOTA models for dense pixel-wise prediction tasks. The proposed VISTA-Net is featured by two aspects: 1) A new structured attention is proposed, which is able to jointly model spatial-level and channel-level dependenc... | SP:5e964be1417deb994f62cd256e24ed7cafd2bd9c |
Manifold Regularization for Locally Stable Deep Neural Networks | 1 INTRODUCTION . Recent results in deep learning highlight the remarkable performance deep neural networks can achieve on tasks using data from the natural world , such as images , video , and audio . Though such data inhabits an input space of high dimensionality , the physical processes which generate the data often ... | This work introduces manifold regularization as an approach for learning stable deep nets, towards the goal of adversarial robustness. Several regularizers are proposed: intrinsic, sparse Laplacian and Hamming regularizers. As the proposed method relies only on adding these regularization terms to the loss, it is more ... | SP:0d62919086db1e43bdd5acbb80c25f82e5466cf6 |
Inverse Constrained Reinforcement Learning | 1 INTRODUCTION . Reward functions are a critical component in reinforcement learning settings . As such , it is important that reward functions are designed accurately and are well-aligned with the intentions of the human designer . This is known as agent ( or value ) alignment ( see , e.g. , Leike et al . ( 2018 ; 201... | The submission focuses on a variant of inverse reinforcement learning, where the learner knows the task reward but is unaware of hard constraints that need to be respected while completing the task. The authors provide an algorithm to recover these constraints from expert demonstrations. The proposed algorithm builds u... | SP:a1bb6da48c8ed54c0bbc88d2109a17276a529c5f |
The Lipschitz Constant of Self-Attention | 1 INTRODUCTION . Lipschitz continuity is a strong form of continuity for functions . Loosely speaking , a function is Lipschitz continuous if changing its input by a certain amount can not change its output by more than K times that amount . The constant K is a hard constraint on how rapidly the function ’ s output can... | This paper studies the Lipschitz continuity properties of self-attention. It is proved that the widely-used dot-product self-attention is not Lipschitz continuous. A novel L2 self-attention is proposed and proven to be Lipschitz continuous. Experiments show that the upper bound of Lipschitz constant for L2 self-attenti... | SP:cd6f5c3ee37991ff572589467b2216ba364275ba |
Robust and Generalizable Visual Representation Learning via Random Convolutions | 1 INTRODUCTION . Generalizability and robustness to out-of-distribution samples have been major pain points when applying deep neural networks ( DNNs ) in real world applications ( Volpi et al. , 2018 ) . Though DNNs are typically trained on datasets with millions of training samples , they still lack robustness to dom... | This paper proposes a simple way to increase the robustness of the learned representations in a network perform a series of object recognition tasks by adding a random convolution layer as a pre-processing stage, thus “filtering the image” and preserving the global shape but altering the local `texture’ of the newly tr... | SP:a0f8915a46a06042331002c9fe2ed47382cc25e9 |
Neural Architecture Search of SPD Manifold Networks | 1 INTRODUCTION . Designing a favorable neural network architecture for a given application requires a lot of time , effort , and domain expertise . To mitigate this issue , researchers in the recent years have started developing algorithms to automate the design process of neural network architectures ( Zoph & Le , 201... | The paper considers a generalization of convolutional neural networks (CNNs) to manifold-valued data such as Symmetric Positive Definite (SPD). This paper proposes a neural architecture search problem of SPD manifold networks. A SPD cell representation and corresponding candidate operation search space is introduced. T... | SP:9d2df0c7b57ce7d4ee0a222ad11361172cb7cbc7 |
Improve Object Detection with Feature-based Knowledge Distillation: Towards Accurate and Efficient Detectors | 1 INTRODUCTION Recently , excellent breakthrough in various domains has been achieved with the success of deep learning ( Ronneberger et al. , 2015 ; Devlin et al. , 2018 ; Ren et al. , 2015 ) . However , the most advanced deep neural networks always consume a large amount of computation and memory , which has limited ... | This paper explores the knowledge distillation problem in object detection. It claims that the failure of knowledge distillation in object detection is mainly caused by the imbalance between pixels of foreground and background, and the relation distillation between different pixels. The authors then propose non-local d... | SP:eb8d96fd5cd18569cfa519c5a09af90ea272d533 |
Learning Structural Edits via Incremental Tree Transformations | 1 INTRODUCTION . Iteratively revising existing data for a certain purpose is ubiquitous . For example , researchers repetitively polish their manuscript until the writing becomes satisfactory ; computer programmers keep editing existing code snippets and fixing bugs until desired programs are produced . Can we properly... | The paper presents an approach for predicting edits in programs, by modeling the programs as trees. The approach is mainly an extension of Yin et al. (2019), with the main difference that the model is required to predict only the output **actions**, instead of generating the entire output tree as in Yin et al. (2019). ... | SP:d5f2c31689e6b6f52bb6f21916e8acacba444f76 |
LowKey: Leveraging Adversarial Attacks to Protect Social Media Users from Facial Recognition | 1 INTRODUCTION . Facial recognition systems ( FR ) are widely deployed for mass surveillance by government agencies , government contractors , and private companies alike on massive databases of images belonging to private individuals ( Hartzog , 2020 ; Derringer , 2019 ; Weise & Singer , 2020 ) . Recently , these syst... | This paper presents a method/tool, i.e., LowKey, to protect user privacy which leverages adversarial attacks to pre-process facial images against the black-box facial recognition system in social media, yet the processed facial images remain visually acceptable. The LowKey method proposes to attack an ensemble of facer... | SP:f3fedc975625ffe149ab536fdf537871fcadcbbf |
Goal-Driven Imitation Learning from Observation by Inferring Goal Proximity | 1 INTRODUCTION . Humans are capable of effectively leveraging demonstrations from experts to solve a variety of tasks . Specifically , by watching others performing a task , we can learn to infer how close we are to completing the task , and then take actions towards states closer to the goal of the task . For example ... | The authors propose a new method for imitation learning from observation that attempts to estimate and leverage a notion of goal proximity in order to help the learning process. The authors provide a framework for computing this estimate, and a technique for using that estimate -- along with a measure of uncertainty --... | SP:174fe6b43a8516e5ea5323ce96a66d64b8745130 |
Adaptive Gradient Methods Can Be Provably Faster than SGD with Random Shuffling | 1 INTRODUCTION . We consider the finite sum minimization problem in stochastic optimization : min x2Rd f ( x ) = 1 n nX i=1 fi ( x ) , ( 1 ) where f is the objective function and its component functions fi : Rd ! R are smooth and possibly non-convex . This formulation has been used extensively in training neural networ... | This paper shows that adaptive learning rates are beneficial for finding critical points of finite-sum optimization problems. In particular, with appropriate learning rates, a variant of adagrad can find a epsilon critical point in \tilde O(1/epsilon^2) iterations. This improves upon previous results of either O(1/\eps... | SP:c0bbcd2b046db616816cf717a30e2547b501378a |
Batch Reinforcement Learning Through Continuation Method | 1 INTRODUCTION . While RL is fundamentally an online learning paradigm , many practical applications of RL algorithms , e.g. , recommender systems [ 5 , 7 ] or autonomous driving [ 36 ] , fall under the batch RL setup . Under this setting , the agent is asked to learn its policy from a fixed set of interactions collect... | The paper extends soft actor-critic (SAC) to the batch RL setting, replacing the policy entropy in the objective function with the KL divergence from the behavioral policy. The temperature parameter tau weighting the reward agains the KL term is annealed towards zero during the optimization process, which corresponds t... | SP:58854dfcef81cc8a791a3b79939046ccfbf9053e |
Shape-Texture Debiased Neural Network Training | 1 INTRODUCTION . It is known that both shape and texture serve as essential cues for object recognition . A decade ago , computer vision researchers had explicitly designed a variety of hand-crafted features , either based on shape ( e.g. , shape context ( Belongie et al. , 2002 ) and inner distance shape context ( Lin... | The authors propose a method to mitigate the bias towards either texture or shape, in convolutional network training. The method follows the idea from Geirhos et al (2019), but use images randomly sampled from the same dataset, instead of style transfer from paintings. Then, depending on a manually selected hyperparame... | SP:d5a996a81845a53ae405b4aac0a9f5342129d43c |
Synthetic Petri Dish: A Novel Surrogate Model for Rapid Architecture Search | 1 INTRODUCTION . The architecture of deep neural networks ( NNs ) is critical to their performance . This fact motivates neural architecture search ( NAS ) , wherein the choice of architecture is often framed as an automated search for effective motifs , i.e . the design of a repeating recurrent cell or activation func... | This paper presents an approach to accelerating NAS with 'petri-dish' networks, which hope to mimic the response of original networks at a fraction of training time cost. The key idea is to evaluate an architectural setting on a miniaturized network as opposed to the original network. With this approach computational e... | SP:ef8a9ec9f2c482ffacdf56b7a36e1fa567b6ba29 |
SaliencyMix: A Saliency Guided Data Augmentation Strategy for Better Regularization | 1 INTRODUCTION . Machine learning has achieved state-of-the-art ( SOTA ) performance in many fields , especially in computer vision tasks . This success can mainly be attributed to the deep architecture of convolutional neural networks ( CNN ) that typically have 10 to 100 millions of learnable parameters . Such a huge... | This paper proposes a new augmentation method based on CutMix. The authors find out that randomly selecting may mix background textures and this will mislead the model. So, they propose to use saliency maps to control the selection of mixed patches, which is called SaliencyMix. This idea seems easy and reasonable, many... | SP:c4bb47d4a04a539331e2ab2ef62b2854804f6a3c |
Efficient estimates of optimal transport via low-dimensional embeddings | 1 Introduction . Optimal Transport metrics ( Kantorovich , 1960 ) or Wasserstein distances , have emerged successfully in the field of machine learning , as outlined in the review by Peyré et al . ( 2017 ) . They provide machinery to lift distances on X to distances over probability distributions in P ( X ) . They hav... | This paper uses the fact that the Wasserstein distance is decreasing under 1-Lipschitz mappings from the ambient space to a feature space in order to propose more robust (to dimensionality ?) estimation of the Wasserstein distance between two probability distributions. A neural network with some sort of weight renormal... | SP:62a3d12370f3248b2283ea33d6767b1c914bcbe2 |
InfoBERT: Improving Robustness of Language Models from An Information Theoretic Perspective | 1 INTRODUCTION . Self-supervised representation learning pre-trains good feature extractors from massive unlabeled data , which show promising transferability to various downstream tasks . Recent success includes large-scale pre-trained language models ( e.g. , BERT , RoBERTa , and GPT-3 ( Devlin et al. , 2019 ; Liu et... | This work (InfoBERT) proposes additional objectives for transformer finetuning to obtain models more robust to adversarial inputs. The authors first propose a mutual information based information bottleneck objective, next the authors propose an adversarial loss inspired method for identifying robust features and a sub... | SP:d06908461594cd2fb28e636fc85b53589a5e1207 |
Language Controls More Than Top-Down Attention: Modulating Bottom-Up Visual Processing with Referring Expressions | 1 INTRODUCTION . As human beings , we can easily understand the surrounding environment with our visual system and interact with each other using language . Since the work of Winograd ( 1972 ) , developing a system that understands human language in a situated environment is one of the long-standing goals of artificial... | This paper concerns the problem of image segmentation from referring expressions. Given an image and a query phrase about a particular object in the image, the goal is to locate the target object as a mask at the pixel level. The basic framework is U-Net, which consists of two branches: an image encoder and segmentatio... | SP:c2c3dfb15f6f05041cbbe6b4542f5dee3eb4e763 |
Acting in Delayed Environments with Non-Stationary Markov Policies | 1 INTRODUCTION . The body of work on reinforcement learning ( RL ) and planning problem setups has grown vast in recent decades . Examples for such distinctions are different objectives and constraints , assumptions on access to the model or logged trajectories , on-policy or off-policy paradigms , etc . ( Puterman , 2... | This paper investigated the problem of learning agents when there are execution delays. The authors (i) used a two-state MDP example to show some equivalence between execution delay and stochasticity of transitions; (ii) analyzed the action aggregation method, which cumulated all the history and then made decisions. Th... | SP:22d8012175584e1a71a2ebc6bb6d3103ff42f87d |
Feature-Robust Optimal Transport for High-Dimensional Data | 1 INTRODUCTION . Optimal transport ( OT ) is a machine learning problem with several applications in the computer vision and natural language processing communities . The applications include Wasserstein distance estimation ( Peyré et al. , 2019 ) , domain adaptation ( Yan et al. , 2018 ) , multitask learning ( Janati... | This work proposes variants of robust OT/p-wasserstein-dist (3)/(4), where the ground cost is in some sense the maximum over costs with (prefixed) groups of features. The motivation is similar to that for feature selection: where perhaps only few of these groups of features are critical/sufficient for OT purposes. So i... | SP:90db8e0421db85e4e43b6fbed1cb68aab5c414e7 |
Distributed Momentum for Byzantine-resilient Stochastic Gradient Descent | 1 Introduction . Stochastic Gradient Descent ( SGD ) is one of the main optimization algorithm used throughout machine learning . Scaling SGD can mean aggregating more but inevitably less well-sanitized data , and distributing the training over several machines , making SGD even more vulnerable to Byzantine faults : co... | The paper describes an approach to counteract Byzantine attacks for distributed stochastic gradient descent by using the momentum of the gradient computed at the workers, which relies only on memory of the previous momentum. This seems to thwart current attacks in the majority of scenarios tested. The theoretical analy... | SP:0b4980e5cb0ed470b3a93111e76fef1035438077 |
Central Server Free Federated Learning over Single-sided Trust Social Networks | 1 INTRODUCTION . Federated learning has been well recognized as a framework able to protect data privacy Konečnỳ et al . ( 2016 ) ; Smith et al . ( 2017a ) ; Yang et al . ( 2019 ) . State-of-the-art federated learning adopts the centralized network architecture where a centralized node collects the gradients sent fro... | The paper proposed Online Push-Sum (OPS) method, which aims at solving decentralized federated optimization problems under a social network scenario where the centralized authority does not exist in a federated learning (FL) system. A social network application scenario is assumed by OPS where the graph is of single-si... | SP:5c55df2071e1ac64d07205434359816ed60f92f9 |
Fundamental Limits and Tradeoffs in Invariant Representation Learning | Many machine learning applications involve learning representations that achieve two competing goals : To maximize information or accuracy with respect to a target while simultaneously maximizing invariance or independence with respect to a subset of features . Typical examples include privacy-preserving learning , dom... | The paper formulates the problems of learning in invariant representations as a min-max game, exploring tradeoffs between accuracy and invariance of these representations via a geometric plane analysis. Specifically. the paper considers both classification (cross entropy loss) and regressions settings (squared loss).Th... | SP:30cbdae14b7b36f103023b56e32c30c8effbf5e6 |
Fast Binarized Neural Network Training with Partial Pre-training | 1 INTRODUCTION . Quantizing neural networks ( Gupta et al. , 2015 ) , constraining weights and activations to take on values within some small fixed set , is a popular set of techniques for reducing the storage ( Han et al. , 2016 ) or compute ( Fromm et al. , 2020 ) requirements of deep neural networks . Weights and a... | The paper suggests a method for training binary neural networks. The proposed method is to partially train with full precision and then continue with binarized training using the straight-through estimator. The method is very simple and there is very limited technical contribution, so in order to be worthy of publicati... | SP:4d39ce0230993594b0fddb3e2655f6f2cfdd308a |
Shuffle to Learn: Self-supervised learning from permutations via differentiable ranking | 1 INTRODUCTION . Supervised learning has achieved important successes on large annotated datasets ( Deng et al. , 2009 ; Amodei et al. , 2016 ) . However , most available data , whether images , audio , or videos are unlabelled . For this reason , pre-training representations in an unsupervised way , with subsequent fi... | This paper presents a self-supervised learning task of shuffling input patches and demanding the network to learn to unshuffle. A related prior work, Noorozi and Favaro (2016) uses a fixed set of permutations to do this task for a given number of patches, and the current paper argues to expand this idea for the full se... | SP:95bf4c07cd691ae29a95ccffa8883f9c92c3eb02 |
Identifying the Sources of Uncertainty in Object Classification | In image-based object classification , the visual appearance of objects determines which class they are assigned to . External variables that are independent of the object , such as the perspective or the lighting conditions , can modify the object ’ s appearance resulting in ambiguous images that lead to misclassifica... | The paper presents an approach that for every object identifies the factors that have a high impact on the models' uncertainty. The approach consists of i) disentangled representations ii) a classifier on the top of the trained representations iii) technique that select dimensions of representation of an objects' (fact... | SP:eed27a2d9c5d77bfc9aacb5d2ca5c7885b2e29f9 |
Conservative Safety Critics for Exploration | 1 INTRODUCTION . Reinforcement learning ( RL ) is a powerful framework for learning-based control because it can enable agents to learn to make decisions automatically through trial and error . However , in the real world , the cost of those trials – and those errors – can be quite high : a quadruped learning to run as... | This paper introduces a method for performing safe exploration in RL. It addresses the problem of ensuring that partially-trained policies do not visit unsafe regions of the state space, while still being exploratory enough to collect useful training experiences. The proposed technique is based on learning conservative... | SP:fa6456aa23ea8635e04081a043a07915b1c0808f |
$\alpha$VIL: Learning to Leverage Auxiliary Tasks for Multitask Learning | 1 INTRODUCTION . In Machine Learning , we often encounter tasks that are at least similar , if not even almost identical . For example , in Computer Vision , multiple datasets might require object segmentation or recognition ( Deng et al. , 2009 ; LeCun et al. , 1998 ; Lin et al. , 2014 ) whereas in Natural Language Pr... | This paper proposes a novel multi-task learning method which adjusts task weights dynamically during training, by exploiting task-specific updates of the model parameters between training epochs. Specifically, the proposed model takes the differences between the model’s parameters before and after the singletask update... | SP:6945d14d266d0ca1c931a55091166604e7984604 |
Learning What To Do by Simulating the Past | 1 INTRODUCTION . As deep learning has become popular , many parts of AI systems that were previously designed by hand have been replaced with learned components . Neural architecture search has automated architecture design ( Zoph & Le , 2017 ; Elsken et al. , 2019 ) , population-based training has automated hyperparam... | This paper introduces an algorithm, called deep reward learning by simulating the past (deep RLSP), that seeks to infer a reward function by looking at states in demonstration data. An example of this described in the paper is an environment with a vase: if demonstration data shows an intact vase in the presence of an ... | SP:e23149389db2c50bba31eacfdef723e015e58386 |
Amortized Conditional Normalized Maximum Likelihood | 1 INTRODUCTION . Current machine learning methods provide unprecedented accuracy across a range of domains , from computer vision to natural language processing . However , in many high-stakes applications , such as medical diagnosis or autonomous driving , rare mistakes can be extremely costly , and thus effective dep... | The paper presents an approach based on conditional normalized maximum likelihood (CNML) for uncertainty estimation, calibration, and out-of-distribution robustness with deep networks. CNML is intractable to compute in general and therefore the authors propose a tractable approximation which uses approximate Bayesian i... | SP:134b968f05fc55567e46428a36359228efa15c85 |
The role of Disentanglement in Generalisation | 1 INTRODUCTION . Generalisation to unseen data has been a key challenge for neural networks since the early days of connectionism , with considerable debate about whether these models can emulate the kinds of behaviours that are present in humans ( McClelland et al. , 1986 ; Fodor & Pylyshyn , 1988 ; Smolensky , 1987 ;... | Learning disentangled representation is often considered an important step to achieve human-like generalization. This paper studies how the degree of disentanglement affects various forms of generalization. Variational autoencoders (VAEs) is trained with different levels of disentanglement on an unsupervised task by e... | SP:81573408426e479610a9d751ebed97dc74f63fb1 |
Long Live the Lottery: The Existence of Winning Tickets in Lifelong Learning | 1 INTRODUCTION . The lottery ticket hypothesis ( LTH ) ( Frankle & Carbin , 2019 ) suggests the existence of an extremely sparse sub-network , within an overparameterized dense neural network , that can reach similar performance as the dense network when trained in isolation with proper initialization . Such a subnetwo... | The research question of this paper is the existence of an extremely sparse network with an initial weight assignment that can be trained online to perform multiple tasks to compete with a dense network, in a lifelong continual learning configuration. Another research question of this paper is how to identify this spar... | SP:f7f0a0d566d1de41a8a8edfed70d363a57c671ef |
Contrasting distinct structured views to learn sentence embeddings | 1 INTRODUCTION . We propose a self-supervised method that builds sentence embeddings from the combination of diverse explicit syntactic structures . Such a method aims at improving the ability of models to perform compositional knowledge . In particular , we evaluate the embedding potential to solve downstream tasks . ... | The authors introduce a pretrianing paradigm based on contrastive learning between multiple syntactic views of the same sentence. The method maximizes representations between different setence encoders when given the same sentence, and minimize the similarity to all other sentence repre sentations. The results on the i... | SP:a74439e1ce3691416cb8557a7662c20855b187ee |
Hamiltonian Q-Learning: Leveraging Importance-sampling for Data Efficient RL | Model-free reinforcement learning ( RL ) , in particular Q-learning is widely used to learn optimal policies for a variety of planning and control problems . However , when the underlying state-transition dynamics are stochastic and high-dimensional , Q-learning requires a large amount of data and incurs a prohibitivel... | This work focuses on dynamic programming in the tabular setting. It proposes to use Hamiltonian Monte-Carlo (HMC) to sample the next states (instead of IID samples) and matrix completion to learn a low-rank Q matrix. It shows theoretical convergence. Experiments on discretized problems (CartPole and an ocean sampling p... | SP:10b339c326238eeef479079dbe713af4ef5b2d92 |
Graph Traversal with Tensor Functionals: A Meta-Algorithm for Scalable Learning | 1 INTRODUCTION . Graph representation learning ( GRL ) has become an invaluable approach for a variety of tasks , such as node classification ( e.g. , in biological and citation networks ; Veličković et al . ( 2018 ) ; Kipf & Welling ( 2017 ) ; Hamilton et al . ( 2017 ) ; Xu et al . ( 2018 ) ) , edge classification (... | The authors propose a "meta-algorithm" for approximating various graph representation learning schemes: generate batches of random trees with fixed fanout (and possibly biased probabilities of selecting different edges), and use them to accumulate information to approximate operations on the graph. The idea is beautif... | SP:eead17fca9c9dd9c1def9e314e19235141fbe709 |
A Truly Constant-time Distribution-aware Negative Sampling | 1 INTRODUCTION . Neural Networks ( NN ) have successfully pushed the boundaries of many application tasks , such as image or text classification ( Wang et al. , 2017 ; Yao et al. , 2019 ) , speech recognition ( Dong et al. , 2018 ) and recommendation systems ( Zhang et al. , 2015 ; Medini et al. , 2019 ) . Many hard AI... | When the number of classes is very large, calculating softmax for classification (e.g., in backpropagation) is computationally costly. Approaches based on negative sampling have been used in literature to alleviate this problem. However, most of existing approaches are (argued to be) either inaccurate or computationall... | SP:1f2d445f78bb495d09e9b796de3662ab6a6b26af |
The Unreasonable Effectiveness of Patches in Deep Convolutional Kernels Methods | 1 INTRODUCTION . Understanding the success of deep convolutional neural networks on images remains challenging because images are high-dimensional signals and deep neural networks are highly-non linear models with a substantial amount of parameters : yet , the curse of dimensionality is seemingly avoided by these model... | This paper proposes a powerful non-learning Kernal based baseline for ImageNet classification. The proposed non-learning Kernal based baseline (which can be interpretable to a vector quantization) shows comparable results (88.5) with AlexNet (89.1) in CIFAR-10 top-1 accuracy. The ImageNet result (39.4) shows that it is... | SP:a0281c7b8cc747c8ced8b7ddfcc56fb6e082eb84 |
Deep Convolution for Irregularly Sampled Temporal Point Clouds | 1 INTRODUCTION . Many real-world problems feature observations that are sparse and irregularly sampled in both space and time . Weather stations scattered across the landscape reporting at variable rates without synchronization ; citizen-science applications producing observations at the whim of individuals ; or even o... | This paper studies the problem of modeling spatial-temporal point clouds which are sampled at irregular space and time points. It proposes the Temporal PointConv model which is an extension of the PointConv model (Wu et al., 2019). In particular, PointConv computes a convolution by aggregating the features of nearby po... | SP:5724a8799e8b77f19887fb7925405a7f151523cc |
AdaLead: A simple and robust adaptive greedy search algorithm for sequence design | 1 INTRODUCTION . An important problem across many domains in biology is the challenge of finding DNA , RNA , or protein sequences which perform a function of interest at a desired level . This task is challenging for two reasons : ( i ) the map φ between sequences X = { x1 , · · · , xn } and their biological function y... | In this work, the authors propose a greedy search approach for designing biological sequences in an active learning, batch setting. The algorithm is a fairly standard evolutionary algorithm which identifies the set of candidates at each iteration by adapting the best candidates from the previous iteration, and then eva... | SP:55e0dbbbabecb54def7b092761ee4e6bd41095c4 |
Zero-Shot Learning with Common Sense Knowledge Graphs | 1 INTRODUCTION . Deep neural networks require large amounts of labeled training data to achieve optimal performance . This is a severe bottleneck , as obtaining large amounts of hand-labeled data is an expensive process . Zero-shot learning is a training strategy which allows a machine learning model to predict novel c... | This paper tackles zero-shot learning by leveraging the large-scale knowledge graph i.e., ConceptNet to propagate the knowledge learned from seen classes to unseen classes. The authors propose a novel propagation rule that aggregates node embeddings by the self-attention technique. It is infeasible to run GCN on such l... | SP:aff048d4b28972615e99e9d2e82258fb0e35f656 |
Blending MPC & Value Function Approximation for Efficient Reinforcement Learning | 1 INTRODUCTION . Model-free Reinforcement Learning ( RL ) is increasingly used in challenging sequential decision-making problems including high-dimensional robotics control tasks ( Haarnoja et al. , 2018 ; Schulman et al. , 2017 ) as well as video and board games ( Silver et al. , 2016 ; 2017 ) . While these approache... | The paper proposes to combine MPC and model-free RL to overcome the possible modelling errors. Thereby the approach achieves the sample-efficiency of MPC and the control quality of model-free RL. The resulting MPQ(\lambda) algorithm uses MPPI to obtain the actions by optimizing the blended MPC objective. The Q-targets ... | SP:42f0f05d335d004a58b91ec986ddd5af72a35a15 |
Multiscale Score Matching for Out-of-Distribution Detection | 1 INTRODUCTION . Modern neural networks do not tend to generalize well to out-of-distribution samples . This phenomenon has been observed in both classifier networks ( Hendrycks & Gimpel ( 2017 ) ; Nguyen et al . ( 2015 ) ; Szegedy et al . ( 2013 ) ) and deep likelihood models ( Nalisnick et al . ( 2018 ) ; Hendrycks e... | The authors leveraged and repurposed Noise Conditioned Score Network (NCSN) that was originally introduced by Song & Ermon (2019) for generative modeling to be used for detection out-of-distribution (OOD) images. The authors unfold the intuition and rationale behind score matching followed by the equivalence of denoisi... | SP:2227006cc52059641d5d7d2fca467d5e392bce65 |
Recurrent Neural Network Architecture based on Dynamic Systems Theory for Data Driven Modelling of Complex Physical Systems | 1 INTRODUCTION . Dynamic systems occur in many different areas of life ( Isermann & Münchhof ( 2011 ) ) . From biology , engineering , medicine to economics and more : Often , if a system changes its state based on a external input , this system can be viewed as a dynamic system . Dynamic system identification is the ... | This paper aims at proposing Dynamic Recurrent Network to understand the underlying system properties of RNNs. By first showing five basic linear transfer functions in dynamic systems theory, the paper formulates DYRNN units. To solve the increasing number of layers issue, they concatenate inputs and intermediate resul... | SP:9f5792697be57f9be662cebfb28e46f123d96682 |
NeurWIN: Neural Whittle Index Network for Restless Bandits via Deep RL | 1 INTRODUCTION . Many sequential decision problems can be modeled as multi-armed bandit problems . A bandit problem models each potential decision as an arm . In each round , we play M arms out of a total of N arms by choosing the corresponding decisions . We then receive a reward from the played arms . The goal is to ... | This paper considers the problem of learning how to control restless bandits. When all parameters of the system are known, Whittle index policy usually offers a good performance. The main contribution of the paper is to propose an algorithm, NeurWIN, that uses a neural network architecture to learn the Whittle indices... | SP:e2d78e6eba2bc0e6273a6ce65549866bc3a29fe7 |
Enforcing robust control guarantees within neural network policies | 1 INTRODUCTION . The field of robust control , dating back many decades , has been able to provide rigorous guarantees on when controllers will succeed or fail in controlling a system of interest . In particular , if the uncertainties in the underlying dynamics can be bounded in specific ways , these techniques can pro... | In this paper, a neural control method is proposed with stability guarantees. The control is assumed to be from a neural network that takes in the state. Stability is guaranteed by projecting the control to the set that satisfies the Lyapunov stability condition for the LQR problem. In particular, minimizing the cost o... | SP:14f1bc469eb56dec5dd691c4a4865aa607fa344e |
Classify and Generate Reciprocally: Simultaneous Positive-Unlabelled Learning and Conditional Generation with Extra Data | 1 INTRODUCTION . Existing machine learning methods , particularly deep learning models , typically require big data to pursue remarkable performance . For instance , conditional deep generative models are able to generate high-fidelity and diverse images , but they have to rely on vast amounts of labeled data ( Lucic e... | This paper proposed the combination of two techniques for improved learning with unlabelled data: 1) Positive-Unlabelled (PU) classifier, and 2) class-conditional GAN (cGAN). The idea is that the PU classifier can help produce more accurate pseudo labels for training of a cGAN, and with the improved cGAN, the generate... | SP:bf9538a602859eaf9e0c3138c5e46c782863a054 |
PODS: Policy Optimization via Differentiable Simulation | 1 INTRODUCTION . The main goal in RL is to formalize principled algorithmic approaches to solving sequential decision-making problems . As a defining characteristic of RL methodologies , agents gain experience by acting in their environments in order to learn how to achieve specific goals . While learning directly in t... | The paper argues for using differentiable simulators for policy optimization. To avoid back propagation through time, the paper splits the policy optimization problem into two steps: i) find improved action sequence for a set of initial conditions, ii) fit a parametric policy to the set of improved action sequences. Fi... | SP:d262c708016b776be1799df31f4b052c107c2b5b |
Outlier Preserving Distribution Mapping Autoencoders | 1 INTRODUCTION . Background . Outlier detection , the task of discovering abnormal instances in a dataset , is critical for applications from fraud detection , error measurement identification to system fault detection ( Singh & Upadhyaya , 2012 ) . Given outliers are by definition rare , it is often infeasible to get ... | The paper proposes an autoencoder-based outlier detection system. The main idea of the paper is to ensure that outlier points are mapped to areas distant from the inliers in the embedding space. To this end, a novel cost function is introduced, which weighs the reconstruction error based on a prior distribution for the... | SP:172fcdf24499acfeba5a1593b48135c0e2b5e6b1 |
Geometry of Program Synthesis | 1 INTRODUCTION . The idea of program synthesis dates back to the birth of modern computation itself ( Turing , 1948 ) and is recognised as one of the most important open problems in computer science ( Gulwani et al. , 2017 ) . However , there appear to be serious obstacles to synthesising programs by gradient descent a... | The authors apply algebraic geometry to program synthesis, by identifying programs with points of analytic varieties. They construct a smooth relaxation to the synthesis problem by considering the space of probability distributions over codes for universal turning machines (which is a smooth/continuous manifold), and t... | SP:d19db4a50cde893b283fb305d8ce11ef37f3edfc |
Network-Agnostic Knowledge Transfer for Medical Image Segmentation | 1 INTRODUCTION . Deep learning often requires a sufficiently large training dataset , which is expensive to build and not easy to share between users . For example , a big challenge with semantic segmentation of medical images is the limited availability of annotated data ( Litjens et al. , 2017 ) . Due to ethical conc... | The paper proposes to use student-teacher training as a way of knowledge transfer between neural networks with different architectures without access to the source data. Instead the authors propose to use a separate dataset to transfer the knowledge of the teacher network and a potential different dataset for fine-tuni... | SP:8ddb96d9abf2c524bd664360a755cbe76703c109 |
Better sampling in explanation methods can prevent dieselgate-like deception | 1 INTRODUCTION . Machine learning models are used in many areas where besides predictive performance , their comprehensibility is also important , e.g. , in healthcare , legal domain , banking , insurance , consultancy , etc . Users in those areas often do not trust a machine learning model if they do not understand wh... | This paper proposes a defense against the adversarial attacks on explanation methods described in Slack et al. (2020). In particular, by using sampling methods that more closely resemble the original data distribution, the authors make it difficult for the Out-of-Distribution detector to successfully discriminate betwe... | SP:8c8d93b1668b5497a4d2b318b6d709f200788262 |
How Does Mixup Help With Robustness and Generalization? | 1 INTRODUCTION . Mixup was introduced by Zhang et al . ( 2018 ) as a data augmentation technique . It has been empirically shown to substantially improve test performance and robustness to adversarial noise of state-of-the-art neural network architectures ( Zhang et al. , 2018 ; Lamb et al. , 2019 ; Thulasidasan et al.... | The paper theoretically studies the beneficial effect of mixup on robustness and generalization of machine models. The mixup loss is rewritten to be the sum of the original empirical loss and a regularization term (plus a high order term). For robustness, the regularization term is proven to be upper bound of first an... | SP:2d40321225b606569305bd303ebad2e1711fd07b |
Deep Clustering and Representation Learning that Preserves Geometric Structures | In this paper , we propose a novel framework for Deep Clustering and multimanifold Representation Learning ( DCRL ) that preserves the geometric structure of data . In the proposed DCRL framework , manifold clustering is done in the latent space guided by a clustering loss . To overcome the problem that clusteringorien... | In this paper, the authors proposes a deep clustering model to enable the clustering and representation learning to favor each other via preserving the geometric structure of data. The proposed DCRL framework integrates an isometric loss for local intra-manifold structure and a ranking loss for global inter-manifold st... | SP:f0fdbe6d66e21168cf3653190ea1f751acf8f2bb |
Disambiguating Symbolic Expressions in Informal Documents | 1 INTRODUCTION . Despite huge advancements in machine learning , the task of understanding informal reasoning is still beyond current methods . In fact , it became commonplace that humans annotate informal documents containing reasoning in many domains , e.g . law ( Libal & Steen , 2020 ) . Reasoning is most visible in... | In more mathematical fields, theorem provers and similar systems can validate claims made about formal systems. However, many research contributions come in the form of papers, and thus they are never validated in this way. Math researchers can express their contributions in a special purpose language to do this, but... | SP:2cf935e397f642fb22a861cb62cb395834eef5b6 |
Localized Meta-Learning: A PAC-Bayes Analysis for Meta-Learning Beyond Global Prior | Meta-learning methods learn the meta-knowledge among various training tasks1 and aim to promote the learning of new tasks under the task similarity assumption.2 Such meta-knowledge is often represented as a fixed distribution ; this , however,3 may be too restrictive to capture various specific task information because... | The paper presents an algorithm for offline meta-learning, where tasks are drawn from a distribution and presented to a learner sequentially, the objective being to use accumulated knowledge in order to facilitate the learning of new tasks. The algorithm is motivated from the PAC-Bayes theory of generalization, extende... | SP:c7ca1f4cc4801fa55cf90d97980f49bf144a1a4c |
Diverse Exploration via InfoMax Options | 1 INTRODUCTION . Abstracting a course of action as a higher-level action , or an option ( Sutton et al. , 1999 ) , is a key ability for reinforcement learning ( RL ) agents in several aspects , including exploration . In RL problems , an agent learns to approximate an optimal policy only from experience , given no prio... | The authors propose a modification of the option-critic algorithm for hierarchical reinforcement learning. The proposed algorithm modifies how the termination conditions of the options are improved by experience. Specifically, the algorithm aims to maximize the mutual information between the options and their terminati... | SP:f24872ae71c6883964f865312805f1c969e97d2c |
Learning Long-term Visual Dynamics with Region Proposal Interaction Networks | 1 INTRODUCTION . As argued by Kenneth Craik , if an organism carries a model of external reality and its own possible actions within its head , it is able to react in a much fuller , safer and more competent manner to emergencies which face it ( Craik , 1952 ) . Indeed , building prediction models has been long studied... | The paper proposes a variation of interaction networks (IN) called region proposal interaction networks. The key idea is to have a richer object-centric feature representation using ROI-Pooling to encode the objects for prediction and use convolution operators to help the IN handle the change in the dimensionality of t... | SP:72e513837413282d60e4e2ab71276a0f7856e87e |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.