NeurIPS 2020 accepted papers on Label-efficient and robust learning
174 of the 1,901 papers accepted at NeurIPS 2020 (9.2%) are on Label-efficient and robust learning, compared with 6.6% at NeurIPS 2019. They fall into 9 topics.
See NeurIPS 2020 in the interactive map
Submitting to NeurIPS? 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 · 76 papers
- Balanced Meta-Softmax for Long-Tailed Visual Recognition (315 citations)
- Understanding Global Feature Contributions With Additive Importance Measures (202 citations)
- The Origins and Prevalence of Texture Bias in Convolutional Neural Networks (117 citations)
- Do Adversarially Robust ImageNet Models Transfer Better? (110 citations)
- Learning from Failure: De-biasing Classifier from Biased Classifier (95 citations)
- How Can I Explain This to You? An Empirical Study of Deep Neural Network Explanation Methods (94 citations)
- Calibrating Deep Neural Networks using Focal Loss (86 citations)
- A Group-Theoretic Framework for Data Augmentation (77 citations)
- Geometric Dataset Distances via Optimal Transport (65 citations)
- Simplify and Robustify Negative Sampling for Implicit Collaborative Filtering (63 citations)
- What shapes feature representations? Exploring datasets, architectures, and training (62 citations)
- Provably Consistent Partial-Label Learning (61 citations)
and 64 more in the interactive map.
Zero-/few-shot and meta-learning · 20 papers
- Interventional Few-Shot Learning (122 citations)
- Pre-training via Paraphrasing (120 citations)
- Bayesian Meta-Learning for the Few-Shot Setting via Deep Kernels (75 citations)
- A causal view of compositional zero-shot recognition (69 citations)
- CrossTransformers: spatially-aware few-shot transfer (57 citations)
- Information Maximization for Few-Shot Learning (48 citations)
- Meta-Learning Requires Meta-Augmentation (46 citations)
- Uncertainty-Aware Learning for Zero-Shot Semantic Segmentation (35 citations)
- Compositional Zero-Shot Learning via Fine-Grained Dense Feature Composition (33 citations)
- MESA: Boost Ensemble Imbalanced Learning with MEta-SAmpler (31 citations)
- Meta-learning from Tasks with Heterogeneous Attribute Spaces (11 citations)
- Modeling Task Effects on Meaning Representation in the Brain via Zero-Shot MEG Prediction (9 citations)
and 8 more in the interactive map.
Domain adaptation and generalization · 20 papers
- Self-paced Contrastive Learning with Hybrid Memory for Domain Adaptive Object Re-ID (338 citations)
- What is being transferred in transfer learning? (188 citations)
- Universal Domain Adaptation through Self Supervision (162 citations)
- Domain Generalization via Entropy Regularization (131 citations)
- Unsupervised Sound Separation Using Mixture Invariant Training (102 citations)
- Domain Generalization for Medical Imaging Classification with Linear-Dependency Regularization (82 citations)
- Adversarial Style Mining for One-Shot Unsupervised Domain Adaptation (76 citations)
- Domain Adaptation with Conditional Distribution Matching and Generalized Label Shift (64 citations)
- Robust Optimal Transport with Applications in Generative Modeling and Domain Adaptation (35 citations)
- Domain Adaptation as a Problem of Inference on Graphical Models (28 citations)
- Learning to Adapt to Evolving Domains (26 citations)
- Transferable Calibration with Lower Bias and Variance in Domain Adaptation (22 citations)
and 8 more in the interactive map.
Self-supervised and contrastive · 17 papers
- Bootstrap Your Own Latent - A New Approach to Self-Supervised Learning (3,384 citations)
- Unsupervised Learning of Visual Features by Contrasting Cluster Assignments (2,015 citations)
- What Makes for Good Views for Contrastive Learning? (640 citations)
- wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations (400 citations)
- Hard Negative Mixing for Contrastive Learning (264 citations)
- 3D Self-Supervised Methods for Medical Imaging (127 citations)
- Supervised Contrastive Learning (113 citations)
- Self-supervised learning through the eyes of a child (71 citations)
- Learning Diverse and Discriminative Representations via the Principle of Maximal Coding Rate Reduction (67 citations)
- Cross-lingual Retrieval for Iterative Self-Supervised Training (65 citations)
- Unsupervised Learning of Dense Visual Representations (50 citations)
- Noise2Same: Optimizing A Self-Supervised Bound for Image Denoising (49 citations)
and 5 more in the interactive map.
Semi-supervised learning · 15 papers
- Unsupervised Data Augmentation for Consistency Training (1,598 citations)
- Big Self-Supervised Models are Strong Semi-Supervised Learners (460 citations)
- Rethinking the Value of Labels for Improving Class-Imbalanced Learning (211 citations)
- VIME: Extending the Success of Self- and Semi-supervised Learning to Tabular Domain (98 citations)
- Distribution Aligning Refinery of Pseudo-label for Imbalanced Semi-supervised Learning (87 citations)
- Contrastive learning of global and local features for medical image segmentation with limited annotations (86 citations)
- Uncertainty-aware Self-training for Few-shot Text Classification (86 citations)
- Towards Interpretable Natural Language Understanding with Explanations as Latent Variables (21 citations)
- Unsupervised Semantic Aggregation and Deformable Template Matching for Semi-Supervised Learning (21 citations)
- Not All Unlabeled Data are Equal: Learning to Weight Data in Semi-supervised Learning (20 citations)
- Auxiliary Task Reweighting for Minimum-data Learning (14 citations)
- Semi-Supervised Partial Label Learning via Confidence-Rated Margin Maximization (12 citations)
and 3 more in the interactive map.
Noisy labels · 10 papers
- Early-Learning Regularization Prevents Memorization of Noisy Labels (257 citations)
- Part-dependent Label Noise: Towards Instance-dependent Label Noise (145 citations)
- Dual T: Reducing Estimation Error for Transition Matrix in Label-noise Learning (104 citations)
- Self-Adaptive Training: beyond Empirical Risk Minimization (87 citations)
- Coresets for Robust Training of Deep Neural Networks against Noisy Labels (46 citations)
- Identifying Mislabeled Data using the Area Under the Margin Ranking (45 citations)
- SuperLoss: A Generic Loss for Robust Curriculum Learning (41 citations)
- A Topological Filter for Learning with Label Noise (12 citations)
- Disentangling Human Error from Ground Truth in Segmentation of Medical Images
- Explicit Regularisation in Gaussian Noise Injections
OOD and anomaly detection · 7 papers
- Energy-based Out-of-distribution Detection (423 citations)
- CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances (266 citations)
- Timeseries Anomaly Detection using Temporal Hierarchical One-Class Network (144 citations)
- On the Value of Out-of-Distribution Testing: An Example of Goodhart's Law (86 citations)
- Understanding Anomaly Detection with Deep Invertible Networks through Hierarchies of Distributions and Features (34 citations)
- OOD-MAML: Meta-Learning for Few-Shot Out-of-Distribution Detection and Classification (33 citations)
- Towards Maximizing the Representation Gap between In-Domain & Out-of-Distribution Examples (19 citations)
Knowledge distillation · 5 papers
- Self-Distillation Amplifies Regularization in Hilbert Space (128 citations)
- Self-Distillation as Instance-Specific Label Smoothing (63 citations)
- Agree to Disagree: Adaptive Ensemble Knowledge Distillation in Gradient Space (58 citations)
- Task-Oriented Feature Distillation (23 citations)
- Knowledge Distillation in Wide Neural Networks: Risk Bound, Data Efficiency and Imperfect Teacher (8 citations)
Test-time adaptation · 4 papers
- Improving robustness against common corruptions by covariate shift adaptation (158 citations)
- Learning Loss for Test-Time Augmentation (38 citations)
- The Cone of Silence: Speech Separation by Localization (29 citations)
- Debugging Tests for Model Explanations (7 citations)
