ICML 2023 accepted papers on Label-efficient and robust learning

Recognition and long-tail · 80 papers

• Robust Speech Recognition via Large-Scale Weak Supervision (1,168 citations)
• Change is Hard: A Closer Look at Subpopulation Shift (12 citations)
• Uncertainty Estimation by Fisher Information-based Evidential Deep Learning (12 citations)
• Discover and Cure: Concept-aware Mitigation of Spurious Correlation (11 citations)
• Effective Neural Topic Modeling with Embedding Clustering Regularization (9 citations)
• A Closer Look at the Intervention Procedure of Concept Bottleneck Models (8 citations)
• Deep Regression Unlearning (8 citations)
• OpenFE: Automated Feature Generation with Expert-level Performance (8 citations)
• Pre-training for Speech Translation: CTC Meets Optimal Transport (7 citations)
• Revisiting Discriminative vs. Generative Classifiers: Theory and Implications (7 citations)
• Interpretable Neural-Symbolic Concept Reasoning (6 citations)
• Learning to Maximize Mutual Information for Dynamic Feature Selection (6 citations)

and 68 more in the interactive map.

Self-supervised and contrastive · 21 papers

• BEATs: Audio Pre-Training with Acoustic Tokenizers (41 citations)
• Contrast with Reconstruct: Contrastive 3D Representation Learning Guided by Generative Pretraining (31 citations)
• Efficient Self-supervised Learning with Contextualized Target Representations for Vision, Speech and Language (18 citations)
• CSP: Self-Supervised Contrastive Spatial Pre-Training for Geospatial-Visual Representations (14 citations)
• RankMe: Assessing the Downstream Performance of Pretrained Self-Supervised Representations by Their Rank (12 citations)
• The SSL Interplay: Augmentations, Inductive Bias, and Generalization (7 citations)
• XTab: Cross-table Pretraining for Tabular Transformers (6 citations)
• On the Power of Foundation Models (5 citations)
• Internet Explorer: Targeted Representation Learning on the Open Web (4 citations)
• Not All Semantics are Created Equal: Contrastive Self-supervised Learning with Automatic Temperature Individualization (4 citations)
• Identifying Interpretable Subspaces in Image Representations (3 citations)
• Learning Instance-Specific Augmentations by Capturing Local Invariances (3 citations)

and 9 more in the interactive map.

Domain adaptation and generalization · 17 papers

• Moderately Distributional Exploration for Domain Generalization (13 citations)
• In Search for a Generalizable Method for Source Free Domain Adaptation (5 citations)
• InfoOT: Information Maximizing Optimal Transport (4 citations)
• On Balancing Bias and Variance in Unsupervised Multi-Source-Free Domain Adaptation (4 citations)
• Drug Discovery under Covariate Shift with Domain-Informed Prior Distributions over Functions (3 citations)
• Model Transferability with Responsive Decision Subjects (3 citations)
• ELSA: Efficient Label Shift Adaptation through the Lens of Semiparametric Models (1 citation)
• Online Prototype Alignment for Few-shot Policy Transfer (1 citation)
• Taxonomy-Structured Domain Adaptation (1 citation)
• Unearthing InSights into Mars: Unsupervised Source Separation with Limited Data (1 citation)
• Demystifying Disagreement-on-the-Line in High Dimensions
• Diversity-enhancing Generative Network for Few-shot Hypothesis Adaptation

and 5 more in the interactive map.

Zero-/few-shot and meta-learning · 16 papers

• Prompting Large Language Model for Machine Translation: A Case Study (69 citations)
• The Unreasonable Effectiveness of Few-shot Learning for Machine Translation (24 citations)
• Tuning Language Models as Training Data Generators for Augmentation-Enhanced Few-Shot Learning (19 citations)
• A Closer Look at Few-shot Classification Again (15 citations)
• CHiLS: Zero-Shot Image Classification with Hierarchical Label Sets (15 citations)
• CLIPood: Generalizing CLIP to Out-of-Distributions (11 citations)
• PromptBoosting: Black-Box Text Classification with Ten Forward Passes (11 citations)
• Towards Omni-generalizable Neural Methods for Vehicle Routing Problems (11 citations)
• A Simple Zero-shot Prompt Weighting Technique to Improve Prompt Ensembling in Text-Image Models (9 citations)
• Evolving Semantic Prototype Improves Generative Zero-Shot Learning (4 citations)
• Generative Causal Representation Learning for Out-of-Distribution Motion Forecasting (3 citations)
• Effective Structured Prompting by Meta-Learning and Representative Verbalizer (2 citations)

and 4 more in the interactive map.

OOD and anomaly detection · 11 papers

• In or Out? Fixing ImageNet Out-of-Distribution Detection Evaluation (11 citations)
• Feed Two Birds with One Scone: Exploiting Wild Data for Both Out-of-Distribution Generalization and Detection (7 citations)
• Fascinating Supervisory Signals and Where to Find Them: Deep Anomaly Detection with Scale Learning (6 citations)
• The Value of Out-of-Distribution Data (3 citations)
• Neural Algorithmic Reasoning with Causal Regularisation (2 citations)
• Concept-based Explanations for Out-of-Distribution Detectors (1 citation)
• Hybrid Energy Based Model in the Feature Space for Out-of-Distribution Detection (1 citation)
• Unleashing Mask: Explore the Intrinsic Out-of-Distribution Detection Capability (1 citation)
• Accuracy on the Curve: On the Nonlinear Correlation of ML Performance Between Data Subpopulations
• Detecting Out-of-distribution Data through In-distribution Class Prior
• Provably Invariant Learning without Domain Information

Test-time adaptation · 11 papers

• AdaNPC: Exploring Non-Parametric Classifier for Test-Time Adaptation (13 citations)
• Data Feedback Loops: Model-driven Amplification of Dataset Biases (11 citations)
• Meta-SAGE: Scale Meta-Learning Scheduled Adaptation with Guided Exploration for Mitigating Scale Shift on Combinatorial Optimization (4 citations)
• On Pitfalls of Test-Time Adaptation (4 citations)
• Test-time Adaptation with Slot-Centric Models (1 citation)
• Do Machine Learning Models Learn Statistical Rules Inferred from Data?
• Target-Aware Generative Augmentations for Single-Shot Adaptation
• Test-Time Style Shifting: Handling Arbitrary Styles in Domain Generalization
• Leveraging Proxy of Training Data for Test-Time Adaptation
• ODS: Test-Time Adaptation in the Presence of Open-World Data Shift
• Statistical Learning under Heterogenous Distribution Shift

Noisy labels · 10 papers

• Identifiability of Label Noise Transition Matrix (6 citations)
• CrossSplit: Mitigating Label Noise Memorization through Data Splitting (2 citations)
• Deep Clustering with Incomplete Noisy Pairwise Annotations: A Geometric Regularization Approach (2 citations)
• Understanding Self-Distillation in the Presence of Label Noise (2 citations)
• Evaluating Unsupervised Denoising Requires Unsupervised Metrics (1 citation)
• Mitigating Memorization of Noisy Labels by Clipping the Model Prediction (1 citation)
• Label Distributionally Robust Losses for Multi-class Classification: Consistency, Robustness and Adaptivity
• When does Privileged information Explain Away Label Noise?
• Delving into Noisy Label Detection with Clean Data
• Which is Better for Learning with Noisy Labels: The Semi-supervised Method or Modeling Label Noise?

Semi-supervised learning · 8 papers

• Are labels informative in semi-supervised learning? Estimating and leveraging the missing-data mechanism. (2 citations)
• NP-SemiSeg: When Neural Processes meet Semi-Supervised Semantic Segmentation (2 citations)
• A Picture of the Space of Typical Learnable Tasks (1 citation)
• Deep Anomaly Detection under Labeling Budget Constraints (1 citation)
• Rethinking Weak Supervision in Helping Contrastive Learning (1 citation)
• Tied-Augment: Controlling Representation Similarity Improves Data Augmentation (1 citation)
• Bidirectional Adaptation for Robust Semi-Supervised Learning with Inconsistent Data Distributions
• Learning the Right Layers a Data-Driven Layer-Aggregation Strategy for Semi-Supervised Learning on Multilayer Graphs

Knowledge distillation · 7 papers

• Knowledge Hypergraph Embedding Meets Relational Algebra (6 citations)
• Learning Lightweight Object Detectors via Multi-Teacher Progressive Distillation (3 citations)
• On the Impact of Knowledge Distillation for Model Interpretability (3 citations)
• Revisiting Data-Free Knowledge Distillation with Poisoned Teachers (3 citations)
• EM-Network: Oracle Guided Self-distillation for Sequence Learning
• Random Teachers are Good Teachers
• MetricGAN-OKD: Multi-Metric Optimization of MetricGAN via Online Knowledge Distillation for Speech Enhancement