ICML 2022 accepted papers on Label-efficient and robust learning
137 of the 1,233 papers accepted at ICML 2022 (11.1%) are on Label-efficient and robust learning, compared with 10.8% at ICML 2021. They fall into 9 topics.
See ICML 2022 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 · 60 papers
- A Consistent and Efficient Evaluation Strategy for Attribution Methods (31 citations)
- How Faithful is your Synthetic Data? Sample-level Metrics for Evaluating and Auditing Generative Models (22 citations)
- RankSim: Ranking Similarity Regularization for Deep Imbalanced Regression (16 citations)
- Gaussian Mixture Variational Autoencoder with Contrastive Learning for Multi-Label Classification (15 citations)
- Open-Sampling: Exploring Out-of-Distribution data for Re-balancing Long-tailed datasets (15 citations)
- Correct-N-Contrast: a Contrastive Approach for Improving Robustness to Spurious Correlations (14 citations)
- MetAug: Contrastive Learning via Meta Feature Augmentation (14 citations)
- Frustratingly Easy Transferability Estimation (11 citations)
- Pure Noise to the Rescue of Insufficient Data: Improving Imbalanced Classification by Training on Random Noise Images (11 citations)
- You Only Cut Once: Boosting Data Augmentation with a Single Cut (11 citations)
- Differentiable Top-k Classification Learning (10 citations)
- Do More Negative Samples Necessarily Hurt In Contrastive Learning? (10 citations)
and 48 more in the interactive map.
Domain adaptation and generalization · 17 papers
- Fishr: Invariant Gradient Variances for Out-of-Distribution Generalization (38 citations)
- A Closer Look at Smoothness in Domain Adversarial Training (27 citations)
- Balancing Discriminability and Transferability for Source-Free Domain Adaptation (24 citations)
- Confidence Score for Source-Free Unsupervised Domain Adaptation (21 citations)
- Connect, Not Collapse: Explaining Contrastive Learning for Unsupervised Domain Adaptation (18 citations)
- Head2Toe: Utilizing Intermediate Representations for Better Transfer Learning (16 citations)
- Robust alignment of cross-session recordings of neural population activity by behaviour via unsupervised domain adaptation (14 citations)
- Generalizing to Evolving Domains with Latent Structure-Aware Sequential Autoencoder (5 citations)
- Understanding Gradual Domain Adaptation: Improved Analysis, Optimal Path and Beyond (5 citations)
- Improving Mini-batch Optimal Transport via Partial Transportation (3 citations)
- Provable Domain Generalization via Invariant-Feature Subspace Recovery (2 citations)
- Active Multi-Task Representation Learning (1 citation)
and 5 more in the interactive map.
Self-supervised and contrastive · 15 papers
- Self-supervised learning with random-projection quantizer for speech recognition (30 citations)
- ContentVec: An Improved Self-Supervised Speech Representation by Disentangling Speakers (24 citations)
- Self-Supervised Models of Audio Effectively Explain Human Cortical Responses to Speech (19 citations)
- Understanding Contrastive Learning Requires Incorporating Inductive Biases (15 citations)
- Adversarial Masking for Self-Supervised Learning (14 citations)
- Perfectly Balanced: Improving Transfer and Robustness of Supervised Contrastive Learning (11 citations)
- Interventional Contrastive Learning with Meta Semantic Regularizer (10 citations)
- Augment with Care: Contrastive Learning for Combinatorial Problems (5 citations)
- On the Difficulty of Defending Self-Supervised Learning against Model Extraction (2 citations)
- Contrastive Learning with Boosted Memorization (1 citation)
- Learning to Separate Voices by Spatial Regions (1 citation)
- On the Surrogate Gap between Contrastive and Supervised Losses (1 citation)
and 3 more in the interactive map.
Noisy labels · 11 papers
- Robust Training under Label Noise by Over-parameterization (26 citations)
- Estimating Instance-dependent Bayes-label Transition Matrix using a Deep Neural Network (17 citations)
- Detecting Corrupted Labels Without Training a Model to Predict (16 citations)
- To Smooth or Not? When Label Smoothing Meets Noisy Labels (15 citations)
- Beyond Images: Label Noise Transition Matrix Estimation for Tasks with Lower-Quality Features (11 citations)
- Investigating Why Contrastive Learning Benefits Robustness against Label Noise (10 citations)
- Robust Fine-Tuning of Deep Neural Networks with Hessian-based Generalization Guarantees (6 citations)
- From Noisy Prediction to True Label: Noisy Prediction Calibration via Generative Model (4 citations)
- Transfer and Marginalize: Explaining Away Label Noise with Privileged Information (2 citations)
- Prototype-Anchored Learning for Learning with Imperfect Annotations (1 citation)
- Robust Meta-learning with Sampling Noise and Label Noise via Eigen-Reptile (1 citation)
OOD and anomaly detection · 10 papers
- Mitigating Neural Network Overconfidence with Logit Normalization (74 citations)
- Out-of-Distribution Detection with Deep Nearest Neighbors (57 citations)
- Latent Outlier Exposure for Anomaly Detection with Contaminated Data (17 citations)
- POEM: Out-of-Distribution Detection with Posterior Sampling (15 citations)
- Training OOD Detectors in their Natural Habitats (15 citations)
- Partial and Asymmetric Contrastive Learning for Out-of-Distribution Detection in Long-Tailed Recognition (10 citations)
- Breaking Down Out-of-Distribution Detection: Many Methods Based on OOD Training Data Estimate a Combination of the Same Core Quantities (5 citations)
- Rich Feature Construction for the Optimization-Generalization Dilemma (5 citations)
- On Distribution Shift in Learning-based Bug Detectors (4 citations)
- FITNESS: (Fine Tune on New and Similar Samples) to detect anomalies in streams with drift and outliers
Zero-/few-shot and meta-learning · 9 papers
- YourTTS: Towards Zero-Shot Multi-Speaker TTS and Zero-Shot Voice Conversion for Everyone (32 citations)
- Channel Importance Matters in Few-Shot Image Classification (25 citations)
- Co-training Improves Prompt-based Learning for Large Language Models (17 citations)
- MAML and ANIL Provably Learn Representations (4 citations)
- Zero-shot AutoML with Pretrained Models (1 citation)
- Attentional Meta-learners for Few-shot Polythetic Classification
- Bayesian Deep Embedding Topic Meta-Learner
- Subspace Learning for Effective Meta-Learning
- The Role of Deconfounding in Meta-learning
Semi-supervised learning · 8 papers
- Active Learning on a Budget: Opposite Strategies Suit High and Low Budgets (33 citations)
- HyperTransformer: Model Generation for Supervised and Semi-Supervised Few-Shot Learning (17 citations)
- NP-Match: When Neural Processes meet Semi-Supervised Learning (6 citations)
- PLATINUM: Semi-Supervised Model Agnostic Meta-Learning using Submodular Mutual Information (2 citations)
- Deep Reference Priors: What is the best way to pretrain a model? (1 citation)
- Auxiliary Learning with Joint Task and Data Scheduling
- Class-Imbalanced Semi-Supervised Learning with Adaptive Thresholding
- Smoothed Adaptive Weighting for Imbalanced Semi-Supervised Learning: Improve Reliability Against Unknown Distribution Data
Test-time adaptation · 4 papers
- Improving Out-of-Distribution Robustness via Selective Augmentation (21 citations)
- Test-Time Training Can Close the Natural Distribution Shift Performance Gap in Deep Learning Based Compressed Sensing (8 citations)
- Minimax Classification under Concept Drift with Multidimensional Adaptation and Performance Guarantees
- Sequential Covariate Shift Detection Using Classifier Two-Sample Tests
