NeurIPS2022 interesting papers

最近NeurIPS放榜了，花了几个小时整理一下自己感兴趣的文章，收集不易，点个赞吧
网站链接：https://nips.cc/Conferences/2022/Schedule?type=Poster

• Differentially Private Model Compression
• Feature Learning in L2-regularized DNNs: Attraction/Repulsion and Sparsity
• REVIVE: Regional Visual Representation Matters in Knowledge-Based Visual Question Answering
• Dataset Inference for Self-Supervised Models
• Neuron with Steady Response Leads to Better Generalization
• Mind the Gap: Understanding the Modality Gap in Multi-modal Contrastive Representation Learning
• Structural Pruning via Latency-Saliency Knapsack
• INRAS: Implicit Neural Representation for Audio Scenes
• What Makes a “Good” Data Augmentation in Knowledge Distillation - A Statistical Perspective
• On Measuring Excess Capacity in Neural Networks
• SIREN: Shaping Representations for OOD Detection
• Distributionally robust weighted k-nearest neighbors
• Effects of Data Geometry in Early Deep Learning
• Understanding Cross-Domain Few-Shot Learning Based on Domain Similarity and Few-Shot Difficulty
• Retaining Knowledge for Learning with Dynamic Definition
• If Influence Functions are the Answer, Then What is the Question?
• Learning sparse features can lead to overfitting in neural networks
• VisFIS: Improved Visual Feature Importance Supervision with Right-for-Right-Reason Objectives
• CS-Shapley: Class-wise Shapley Values for Data Valuation in Classification
• ‘Why Not Other Classes?”: Towards Class-Contrastive Back-Propagation Explanations
• Self-Supervised Fair Representation Learning without Demographics
• Fairness without Demographics through Knowledge Distillation
• Towards Understanding the Condensation of Neural Networks at Initial Training
• Hub-Pathway: Transfer Learning from A Hub of Pre-trained Models
• Neural Temporal Walks: Motif-Aware Representation Learning on Continuous-Time Dynamic Graphs
• On Robust Multiclass Learnability
• Neural Matching Fields: Implicit Representation of Matching Cost for Semantic Correspondence
• Rethinking and Scaling Up Graph Contrastive Learning: An Extremely Efficient Approach with Group Discrimination
• Implicit Neural Representations with Levels-of-Experts
• Learning to Re-weight Examples with Optimal Transport for Imbalanced Classification
• A Data-Augmentation Is Worth A Thousand Samples
• Exploring Example Influence in Continual Learning
• Bridge the Gap Between Architecture Spaces via A Cross-Domain Predictor
• Deconfounded Representation Similarity for Comparison of Neural Networks
• Training with More Confidence: Mitigating Injected and Natural Backdoors During Training
• Understanding Neural Architecture Search: Convergence and Generalization
• Understanding Robust Learning through the Lens of Representation Similarities
• Efficient Dataset Distillation using Random Feature Approximation
• Distilling Representations from GAN Generator via Squeeze and Span
• AttCAT: Explaining Transformers via Attentive Class Activation Tokens
• On the Safety of Interpretable Machine Learning: A Maximum Deviation Approach
• Task Discovery: Finding the Tasks that Neural Networks Generalize on
• Federated Learning from Pre-Trained Models: A Contrastive Learning Approach
• Improving Self-Supervised Learning by Characterizing Idealized Representations
• Teach Less, Learn More: On the Undistillable Classes in Knowledge Distillation
• Domain Generalization by Learning and Removing Domain-specific Features
• Do We Really Need a Learnable Classifier at the End of Deep Neural Network?
• Pruning has a disparate impact on model accuracy
• Neural network architecture beyond width and depth
• Explaining Graph Neural Networks with Structure-Aware Cooperative Games
• TA-GATES: An Encoding Scheme for Neural Network Architectures
• Is this the Right Neighborhood? Accurate and Query Efficient Model Agnostic Explanations
• Respecting Transfer Gap in Knowledge Distillation
• Transferring Pre-trained Multimodal Representations with Cross-modal Similarity Matching
• Transfer Learning on Heterogeneous Feature Spaces for Treatment Effects Estimation
• Concept Activation Regions: A Generalized Framework For Concept-Based Explanations
• 3DB: A Framework for Debugging Computer Vision Models
• Redundant representations help generalization in wide neural networks
• What You See is What You Classify: Black Box Attributions
• Efficient identification of informative features in simulation-based inference
• Best of Both Worlds Model Selection
• Understanding Self-Supervised Graph Representation Learning from a Data-Centric Perspective
• Lost in Latent Space: Examining failures of disentangled models at combinatorial generalisation
• Does GNN Pretraining Help Molecular Representation?
• Quality Not Quantity: On the Interaction between Dataset Design and Robustness of CLIP
• GlanceNets: Interpretabile, Leak-proof Concept-based Models
• Evolution of Neural Tangent Kernels under Benign and Adversarial Training
• Learning to Scaffold: Optimizing Model Explanations for Teaching
• Dataset Distillation using Neural Feature Regression
• Where do Models go Wrong? Parameter-Space Saliency Maps for Explainability
• Palm up: Playing in the Latent Manifold for Unsupervised Pretraining
• Spherization Layer: Representation Using Only Angles
• Self-Similarity Priors: Neural Collages as Differentiable Fractal Representations
• Pre-Trained Model Reusability Evaluation for Small-Data Transfer Learning
• Task-Agnostic Graph Explanations
• On Neural Network Pruning’s Effect on Generalization
• In What Ways Are Deep Neural Networks Invariant and How Should We Measure This?
• On the Symmetries of Deep Learning Models and their Internal Representations
• Training Subset Selection for Weak Supervision
• Fair Infinitesimal Jackknife: Mitigating the Influence of Biased Training Data Points Without Refitting
• Beyond Separability: Analyzing the Linear Transferability of Contrastive Representations to Related Subpopulations
• Meta-learning for Feature Selection with Hilbert-Schmidt Independence Criterion
• GULP: a prediction-based metric between representations
• Interpreting Operation Selection in Differentiable Architecture Search: A Perspective from Influence-Directed Explanations
• Seeing the forest and the tree: Building representations of both individual and collective dynamics with transformers
• Insights into Pre-training via Simpler Synthetic Tasks
• Orient: Submodular Mutual Information Measures for Data Subset Selection under Distribution Shift
• Knowledge Distillation: Bad Models Can Be Good Role Models
• Listen to Interpret: Post-hoc Interpretability for Audio Networks with NMF
• Weakly Supervised Representation Learning with Sparse Perturbations
• Which Explanation Should I Choose? A Function Approximation Perspective to Characterizing Post hoc Explanations
• FedAvg with Fine Tuning: Local Updates Lead to Representation Learning
• Pruning Neural Networks via Coresets and Convex Geometry: Towards No Assumptions
• Efficient Architecture Search for Diverse Tasks
• Learning to Reason with Neural Networks: Generalization, Unseen Data and Boolean Measures
• Reconstructing Training Data From Trained Neural Networks
• Procedural Image Programs for Representation Learning
• Where to Pay Attention in Sparse Training for Feature Selection?
• Could Giant Pre-trained Image Models Extract Universal Representations?
• On the Interpretability of Regularisation for Neural Networks Through Model Gradient Similarity
• Vision GNN: An Image is Worth Graph of Nodes
• CLEAR: Generative Counterfactual Explanations on Graphs
• Neural Basis Models for Interpretability
• Exploring Linear Feature Scalability of Vision Transformer for Parameter-efficient Fine-tuning
• Private Estimation with Public Data
• Robust Testing in High-Dimensional Sparse Models
• Dataset Factorization for Condensation
• One Layer is All You Need
• Improved Fine-Tuning by Better Leveraging Pre-Training Data
• Don’t Throw Your Model Checkpoints Away
• Finding Differences Between Transformers and ConvNets Using Counterfactual Simulation Testing