ICML 2025 accepted papers on Label-efficient and robust learning

Recognition and long-tail · 114 papers

• Position: Deep Learning is Not So Mysterious or Different (2 citations)
• Should Decision-Makers Reveal Classifiers in Online Strategic Classification? (2 citations)
• TabPFN Unleashed: A Scalable and Effective Solution to Tabular Classification Problems (2 citations)
• ActionPiece: Contextually Tokenizing Action Sequences for Generative Recommendation (1 citation)
• ADIOS: Antibody Development via Opponent Shaping (1 citation)
• Archetypal SAE: Adaptive and Stable Dictionary Learning for Concept Extraction in Large Vision Models (1 citation)
• BCE vs. CE in Deep Feature Learning (1 citation)
• Editable Concept Bottleneck Models (1 citation)
• Features are fate: a theory of transfer learning in high-dimensional regression (1 citation)
• "Who experiences large model decay and why?" A Hierarchical Framework for Diagnosing Heterogeneous Performance Drift
• A Mixed-Curvature based Pre-training Paradigm for Multi-Task Vehicle Routing Solver
• A Variational Framework for Improving Naturalness in Generative Spoken Language Models

and 102 more in the interactive map.

Self-supervised and contrastive · 21 papers

• The Brain's Bitter Lesson: Scaling Speech Decoding With Self-Supervised Learning (1 citation)
• An Augmentation-Aware Theory for Self-Supervised Contrastive Learning
• Clustering Properties of Self-Supervised Learning
• Clustering via Self-Supervised Diffusion
• Collapse-Proof Non-Contrastive Self-Supervised Learning
• Contextures: Representations from Contexts
• How does Labeling Error Impact Contrastive Learning? A Perspective from Data Dimensionality Reduction
• Objective drives the consistency of representational similarity across datasets
• On the Importance of Embedding Norms in Self-Supervised Learning
• On the Out-of-Distribution Generalization of Self-Supervised Learning
• Position: An Empirically Grounded Identifiability Theory Will Accelerate Self Supervised Learning Research
• Robust Autonomy Emerges from Self-Play

and 9 more in the interactive map.

Domain adaptation and generalization · 20 papers

• Concept-Based Unsupervised Domain Adaptation (1 citation)
• MaskTwins: Dual-form Complementary Masking for Domain-Adaptive Image Segmentation (1 citation)
• A General Representation-Based Approach to Multi-Source Domain Adaptation
• Adapter Naturally Serves as Decoupler for Cross-Domain Few-Shot Semantic Segmentation
• Balanced Learning for Domain Adaptive Semantic Segmentation
• Chameleon: A Flexible Data-mixing Framework for Language Model Pretraining and Finetuning
• COSDA: Counterfactual-based Susceptibility Risk Framework for Open-Set Domain Adaptation
• Domain2Vec: Vectorizing Datasets to Find the Optimal Data Mixture without Training
• LangDAug: Langevin Data Augmentation for Multi-Source Domain Generalization in Medical Image Segmentation
• Merge-Friendly Post-Training Quantization for Multi-Target Domain Adaptation
• Random Registers for Cross-Domain Few-Shot Learning
• Revisiting Continuity of Image Tokens for Cross-domain Few-shot Learning

and 8 more in the interactive map.

OOD and anomaly detection · 17 papers

• Adaptive Multi-prompt Contrastive Network for Few-shot Out-of-distribution Detection
• Avoiding Leakage Poisoning: Concept Interventions Under Distribution Shifts
• Controlling Neural Collapse Enhances Out-of-Distribution Detection and Transfer Learning
• CostFilter-AD: Enhancing Anomaly Detection through Matching Cost Filtering
• Improving Out-of-Distribution Detection via Dynamic Covariance Calibration
• Improving Out-of-Distribution Detection with Markov Logic Networks
• LAION-C: An Out-of-Distribution Benchmark for Web-Scale Vision Models
• OOD-Chameleon: Is Algorithm Selection for OOD Generalization Learnable?
• Towards a Unified Framework of Clustering-based Anomaly Detection
• A Variational Information Theoretic Approach to Out-of-Distribution Detection
• Demeaned Sparse: Efficient Anomaly Detection by Residual Estimate
• Mahalanobis++: Improving OOD Detection via Feature Normalization

and 5 more in the interactive map.

Knowledge distillation · 15 papers

• Pre-Training Graph Contrastive Masked Autoencoders are Strong Distillers for EEG (1 citation)
• Discrepancies are Virtue: Weak-to-Strong Generalization through Lens of Intrinsic Dimension
• Distillation Scaling Laws
• Distilling the Knowledge in Data Pruning
• FEAT-KD: Learning Concise Representations for Single and Multi-Target Regression via TabNet Knowledge Distillation
• Weak-to-Strong Generalization Even in Random Feature Networks, Provably
• When Data-Free Knowledge Distillation Meets Non-Transferable Teacher: Escaping Out-of-Distribution Trap is All You Need
• ABKD: Pursuing a Proper Allocation of the Probability Mass in Knowledge Distillation via $\alpha$-$\beta$-Divergence
• Asymmetric Decision-Making in Online Knowledge Distillation: Unifying Consensus and Divergence
• DA-KD: Difficulty-Aware Knowledge Distillation for Efficient Large Language Models
• Enhancing Logits Distillation with Plug&Play Kendall's $\tau$ Ranking Loss
• Ensemble Distribution Distillation via Flow Matching

and 3 more in the interactive map.

Zero-/few-shot and meta-learning · 12 papers

• PertEval-scFM: Benchmarking Single-Cell Foundation Models for Perturbation Effect Prediction (40 citations)
• GenZSL: Generative Zero-Shot Learning Via Inductive Variational Autoencoder (1 citation)
• Provably Improving Generalization of Few-shot models with Synthetic Data (1 citation)
• Zero-shot Meta-learning for Tabular Prediction Tasks with Adversarially Pre-trained Transformer (1 citation)
• A Generalization Theory for Zero-Shot Prediction
• InfoSAM: Fine-Tuning the Segment Anything Model from An Information-Theoretic Perspective
• Learning from True-False Labels via Multi-modal Prompt Retrieving
• Self-cross Feature based Spiking Neural Networks for Efficient Few-shot Learning
• BounDr.E: Predicting Drug-likeness via Biomedical Knowledge Alignment and EM-like One-Class Boundary Optimization
• Discovering a Zero (Zero-Vector Class of Machine Learning)
• Do Not Mimic My Voice : Speaker Identity Unlearning for Zero-Shot Text-to-Speech
• Universal Biological Sequence Reranking for Improved De Novo Peptide Sequencing

Semi-supervised learning · 9 papers

• GraphCL: Graph-based Clustering for Semi-Supervised Medical Image Segmentation (1 citation)
• A Square Peg in a Square Hole: Meta-Expert for Long-Tailed Semi-Supervised Learning
• Info-Coevolution: An Efficient Framework for Data Model Coevolution
• Towards the Efficient Inference by Incorporating Automated Computational Phenotypes under Covariate Shift
• Unsupervised Learning for Class Distribution Mismatch
• Beyond Confidence: Exploiting Homogeneous Pattern for Semi-Supervised Semantic Segmentation
• Non-Stationary Predictions May Be More Informative: Exploring Pseudo-Labels with a Two-Phase Pattern of Training Dynamics
• Rethinking Confidence Scores and Thresholds in Pseudolabeling-based SSL
• Semi-Supervised Blind Quality Assessment with Confidence-quantifiable Pseudo-label Learning for Authentic Images

Test-time adaptation · 9 papers

• Test-Time Adaptation with Binary Feedback (1 citation)
• WATCH: Adaptive Monitoring for AI Deployments via Weighted-Conformal Martingales (1 citation)
• Beyond Entropy: Region Confidence Proxy for Wild Test-Time Adaptation
• IT$^3$: Idempotent Test-Time Training
• Drug-TTA: Test-Time Adaptation for Drug Virtual Screening via Multi-task Meta-Auxiliary Learning
• PTTA: Purifying Malicious Samples for Test-Time Model Adaptation
• Test-time Adaptation on Graphs via Adaptive Subgraph-based Selection and Regularized Prototypes
• Test-time Correlation Alignment
• Test-Time Selective Adaptation for Uni-Modal Distribution Shift in Multi-Modal Data

Noisy labels · 9 papers

• Improved Algorithm for Deep Active Learning under Imbalance via Optimal Separation (2 citations)
• Automatically Identify and Rectify: Robust Deep Contrastive Multi-view Clustering in Noisy Scenarios
• Bipartite Ranking From Multiple Labels: On Loss Versus Label Aggregation
• On the Role of Label Noise in the Feature Learning Process
• Rethinking Benign Overfitting in Two-Layer Neural Networks
• Retraining with Predicted Hard Labels Provably Increases Model Accuracy
• Confidence Difference Reflects Various Supervised Signals in Confidence-Difference Classification
• Learning Imbalanced Data with Beneficial Label Noise
• MOGIC: Metadata-infused Oracle Guidance for Improved Extreme Classification