ICML 2021 accepted papers on Label-efficient and robust learning
128 of the 1,184 papers accepted at ICML 2021 (10.8%) are on Label-efficient and robust learning, compared with 10.0% at ICML 2020. They fall into 9 topics.
See ICML 2021 in the interactive map
Submitting to ICML? Checklist →
Working on something in this area? Describe your idea in scime atlas to see the closest papers by meaning and how crowded the spot has become.
Topics and papers
Recognition and long-tail · 61 papers
- Delving into Deep Imbalanced Regression (120 citations)
- Just Train Twice: Improving Group Robustness without Training Group Information (69 citations)
- On Disentangled Representations Learned from Correlated Data (47 citations)
- Order-Agnostic Cross Entropy for Non-Autoregressive Machine Translation (46 citations)
- Backdoor Scanning for Deep Neural Networks through K-Arm Optimization (45 citations)
- LogME: Practical Assessment of Pre-trained Models for Transfer Learning (41 citations)
- LARNet: Lie Algebra Residual Network for Face Recognition (28 citations)
- Break-It-Fix-It: Unsupervised Learning for Program Repair (27 citations)
- Characterizing Structural Regularities of Labeled Data in Overparameterized Models (25 citations)
- ASAM: Adaptive Sharpness-Aware Minimization for Scale-Invariant Learning of Deep Neural Networks (19 citations)
- Active Deep Probabilistic Subsampling (18 citations)
- Active Testing: Sample-Efficient Model Evaluation (16 citations)
and 49 more in the interactive map.
Self-supervised and contrastive · 15 papers
- Barlow Twins: Self-Supervised Learning via Redundancy Reduction (770 citations)
- Whitening for Self-Supervised Representation Learning (42 citations)
- UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data (39 citations)
- Understanding self-supervised learning dynamics without contrastive pairs (33 citations)
- CLOCS: Contrastive Learning of Cardiac Signals Across Space, Time, and Patients (29 citations)
- Contrastive Learning Inverts the Data Generating Process (27 citations)
- Self-Damaging Contrastive Learning (14 citations)
- Toward Understanding the Feature Learning Process of Self-supervised Contrastive Learning (11 citations)
- Towards Domain-Agnostic Contrastive Learning (9 citations)
- Self-supervised and Supervised Joint Training for Resource-rich Machine Translation (8 citations)
- Function Contrastive Learning of Transferable Meta-Representations (4 citations)
- GeomCA: Geometric Evaluation of Data Representations (2 citations)
and 3 more in the interactive map.
Noisy labels · 13 papers
- Confidence Scores Make Instance-dependent Label-noise Learning Possible (35 citations)
- Provably End-to-end Label-noise Learning without Anchor Points (26 citations)
- Learning Noise Transition Matrix from Only Noisy Labels via Total Variation Regularization (24 citations)
- Clusterability as an Alternative to Anchor Points When Learning with Noisy Labels (21 citations)
- Class2Simi: A Noise Reduction Perspective on Learning with Noisy Labels (20 citations)
- Asymmetric Loss Functions for Learning with Noisy Labels (12 citations)
- Understanding Instance-Level Label Noise: Disparate Impacts and Treatments (8 citations)
- Pointwise Binary Classification with Pairwise Confidence Comparisons (3 citations)
- Locally Adaptive Label Smoothing Improves Predictive Churn (2 citations)
- RATT: Leveraging Unlabeled Data to Guarantee Generalization (2 citations)
- Label Inference Attacks from Log-loss Scores (1 citation)
- Wasserstein Distributional Normalization For Robust Distributional Certification of Noisy Labeled Data (1 citation)
and 1 more in the interactive map.
Domain adaptation and generalization · 12 papers
- Domain Generalization using Causal Matching (75 citations)
- Environment Inference for Invariant Learning (52 citations)
- KD3A: Unsupervised Multi-Source Decentralized Domain Adaptation via Knowledge Distillation (46 citations)
- Representation Subspace Distance for Domain Adaptation Regression (29 citations)
- Selecting Data Augmentation for Simulating Interventions (21 citations)
- LAMDA: Label Matching Deep Domain Adaptation (13 citations)
- A Theory of Label Propagation for Subpopulation Shift (8 citations)
- f-Domain Adversarial Learning: Theory and Algorithms (8 citations)
- Aggregating From Multiple Target-Shifted Sources (7 citations)
- A Discriminative Technique for Multiple-Source Adaptation (6 citations)
- Sequential Domain Adaptation by Synthesizing Distributionally Robust Experts (1 citation)
- CARTL: Cooperative Adversarially-Robust Transfer Learning
Zero-/few-shot and meta-learning · 10 papers
- Data Augmentation for Meta-Learning (26 citations)
- Improving Generalization in Meta-learning via Task Augmentation (19 citations)
- Learning a Universal Template for Few-shot Dataset Generalization (9 citations)
- A large-scale benchmark for few-shot program induction and synthesis (8 citations)
- Parameterless Transductive Feature Re-representation for Few-Shot Learning (7 citations)
- Bridging Multi-Task Learning and Meta-Learning: Towards Efficient Training and Effective Adaptation (6 citations)
- Unsupervised Embedding Adaptation via Early-Stage Feature Reconstruction for Few-Shot Classification (6 citations)
- A Representation Learning Perspective on the Importance of Train-Validation Splitting in Meta-Learning (4 citations)
- Few-Shot Conformal Prediction with Auxiliary Tasks (3 citations)
- Model Fusion for Personalized Learning
OOD and anomaly detection · 6 papers
- Accuracy on the Line: on the Strong Correlation Between Out-of-Distribution and In-Distribution Generalization (30 citations)
- Neural Transformation Learning for Deep Anomaly Detection Beyond Images (22 citations)
- Can Subnetwork Structure Be the Key to Out-of-Distribution Generalization? (15 citations)
- Non-Negative Bregman Divergence Minimization for Deep Direct Density Ratio Estimation (6 citations)
- DriftSurf: Stable-State / Reactive-State Learning under Concept Drift (4 citations)
- Heterogeneous Risk Minimization (2 citations)
Semi-supervised learning · 5 papers
- Dash: Semi-Supervised Learning with Dynamic Thresholding (53 citations)
- Sinkhorn Label Allocation: Semi-Supervised Classification via Annealed Self-Training (7 citations)
- Message Passing Adaptive Resonance Theory for Online Active Semi-supervised Learning (3 citations)
- Learning from Biased Data: A Semi-Parametric Approach (1 citation)
- Self-Tuning for Data-Efficient Deep Learning
Knowledge distillation · 4 papers
- Zero-Shot Knowledge Distillation from a Decision-Based Black-Box Model (14 citations)
- A statistical perspective on distillation (8 citations)
- Simultaneous Similarity-based Self-Distillation for Deep Metric Learning (4 citations)
- Interpreting and Disentangling Feature Components of Various Complexity from DNNs (2 citations)
Test-time adaptation · 2 papers
- WILDS: A Benchmark of in-the-Wild Distribution Shifts (278 citations)
- Out-of-Distribution Generalization via Risk Extrapolation (REx) (251 citations)
