ICML-21 待读的 Paper

2021.6.3

ICML 2021官方发布接收论文，共有5513篇论文投稿，共有1184篇接受（包括1018篇短论文和166篇长论文），接受率21.48%。

具体list 见： ICML-21 Accepted paper list

Interesting paper

• A statistical perspective on distillation

• Model Fusion for Personalized Learning

• Learning Bounds for Open-Set Learning

• Learning from the Crowd with Pairwise Comparison

• Generalization Bounds in the Presence of Outliers: a Median-of-Means Study

• SiameseXML: Siamese Networks meet Extreme Classifiers with 100M Labels

• Progressive Learning for Convolutional Neural Networks

• On the price of explainability for some clustering problems

• Learning Curves for Analysis of Deep Networks

• Self-Tuning for Data-Efficient Deep Learning

• Adversarial robustness guarantees for random deep neural networks

• AutoSampling: Search for Effective Data Sampling Schedules

• Soft then Hard: Rethinking the Quantization in Neural Image Compression

• Implicit Bias of Linear RNNs

• Break-It-Fix-It: Learning to Repair Code from Unlabeled Data

• Classification with Rejection Based on Cost-sensitive Classification

• Oblivious Sketching for Logistic Regression

• One Pass Late Fusion Multi-view Clustering

• Exact Gap between Generalization Error and Uniform Convergence in Random Feature Models

• Attention is not all you need: pure attention loses rank doubly exponentially with depth

• Pointwise Binary Classification with Pairwise Confidence Comparisons (Feng Lei 这个人去了重庆大学）

• Learning from Similarity-Confidence Data

• Towards Understanding Learning in Neural Networks with Linear Teachers

• Leveraged Weighted Loss for Partial Label Learning

• Active Testing: Sample-Efficient Model Evaluation

• Robust Unsupervised Learning via L-statistic Minimization

• RATT: Leveraging Unlabeled Data to Guarantee Generalization

• Sharper Generalization Bounds for Clustering

• Towards Better Robust Generalization with Shift Consistency Regularization

• Dash: Semi-Supervised Learning with Dynamic Thresholding

• Sinkhorn Label Allocation: Semi-Supervised Classification via Annealed Self-Training

• Locally Adaptive Label Smoothing Improves Predictive Churn

Noisy labels:

• Lower-bounded proper losses for weakly supervised classification

• Disambiguation of Weak Supervision leading to Exponential Convergence rates

• Clusterability as an Alternative to Anchor Points When Learning with Noisy Labels

• Label Distribution Learning Machine

• Discriminative Complementary-Label Learning with Weighted Loss

• Multi-Dimensional Classification via Sparse Label Encoding

• On the Inherent Regularization Effects of Noise Injection During Training

• Provably End-to-end Label-noise Learning without Anchor Points

• Improved OOD Generalization via Adversarial Training and Pretraing

• Can Subnetwork Structure Be the Key to Out-of-Distribution Generalization?

• A General Framework For Detecting Anomalous Inputs to DNN Classifiers

• Accuracy on the Line: on the Strong Correlation Between Out-of-Distribution and In-Distribution Generalization

• The importance of understanding instance-level noisy labels

• Confidence Scores Make Instance-dependent Label-noise Learning Possible

• Learning Noise Transition Matrix from Only Noisy Labels via Total Variation Regularization

• Wasserstein Distributional Normalization For Robust Distributional Certification of Noisy Labeled Data

• Understanding and Mitigating Accuracy Disparity in Regression

• Revealing the Structure of Deep Neural Networks via Convex Duality

• Learning from Biased Data: A Semi-Parametric Approach

• Class2Simi: A Noise Reduction Perspective on Learning with Noisy Labels

• Learning Deep Neural Networks under Agnostic Corrupted Supervision

• Learning from Noisy Labels with No Change to the Training Process

• Adversarial Multi Class Learning under Weak Supervision with Performance Guarantees

OOD：

• Amortized Conditional Normalized Maximum Likelihood: Reliable Out of Distribution Uncertainty Estimation
• Delving into Deep Imbalanced Regression
• Matrix Sketching for Secure Collaborative Machine Learning
• A Collective Learning Framework to Boost GNN Expressiveness for Node Classification
• Out-of-Distribution Generalization via Risk Extrapolation (REx)
• Don’t Just Blame Over-parametrization for Over-confidence: Theoretical Analysis of Calibration in Binary Classification
• Graph Convolution for Semi-Supervised Classification: Improved Linear Separability and Out-of-Distribution Generalization
• Failure Modes and Opportunities in Out-of-distribution Detection with Deep Generative Models

GNN

• On Explainability of Graph Neural Networks via Subgraph Explorations
• GRAND: Graph Neural Diffusion
• Optimization of Graph Neural Networks: Implicit Acceleration by Skip Connections and More Depth
• Information Obfuscation of Graph Neural Networks
• Generative Causal Explanations for Graph Neural Networks
• How Framelets Enhance Graph Neural Networks
• GraphNorm: A Principled Approach to Accelerating Graph Neural Network Training
• Let's Agree to Degree: Comparing Graph Convolutional Networks in the Message-Passing Framework
• Memory-Efficient Graph Neural Networks
• A Unified Lottery Ticket Hypothesis for Graph Neural Networks
• Directional Graph Networks
• Graph Contrastive Learning Automated
• Automated Graph Representation Learning with Hyperparameter Importance Explanation
• E(n) Equivariant Graph Neural Networks
• Breaking the Limits of Message Passing Graph Neural Networks
• DeepWalking Backwards: From Embeddings Back to Graphs
• Elastic Graph Neural Networks
• Graph Neural Networks Inspired by Classical Iterative Algorithms

Contrastive learning

• Large-Margin Contrastive Learning with Distance Polarization Regularizer

• CLOCS: Contrastive Learning of Cardiac Signals Across Space, Time, and Patients

• Self-supervised Graph-level Representation Learning with Local and Global Structure

• Towards Domain-Agnostic Contrastive Learning

• Unsupervised Representation Learning via Neural Activation Coding

• Whitening for Self-Supervised Representation Learning

• Barlow Twins: Self-Supervised Learning via Redundancy Reduction

• Self-Damaging Contrastive Learning

• Contrastive Learning Inverts the Data Generating Process

• Dissecting Supervised Constrastive Learning

• Neighborhood Contrastive Learning Applied to Online Patient Monitoring

• Toward Understanding the Feature Learning Process of Self-supervised Contrastive Learning

• Function Contrastive Learning of Transferable Meta-Representations

• Understanding self-supervised learning dynamics without contrastive pairs

• ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision

推荐搜索

• Rethinking Neural vs. Matrix-Factorization Collaborative Filtering: the Theoretical Perspectives
• Meta Latents Learning for Open-World Recommender Systems
• Learning Self-Modulating Attention in Continuous Time Space with Applications to Sequential Recommendation
• Quantifying Availability and Discovery in Recommender Systems via Stochastic Reachability
• Correcting Exposure Bias for Link Recommendation
• Estimating α-Rank from A Few Entries with Low Rank Matrix Completion
• Quantifying Availability and Discovery in Recommender Systems via Stochastic Reachability
• Follow-the-Regularizer-Leader Routes to Chaos in Routing Games
• Matrix Completion with Model-free Weighting
• Correcting Exposure Bias for Link Recommendation
  Meta Latents Learning for Open-World Recommender Systems

Oops!

• LAMDA: Label Matching Deep Domain Adaptation
• Making Paper Reviewing Robust to Bid Manipulation Attacks