ICLR 2020 accepted papers on Label-efficient and robust learning
79 of the 687 papers accepted at ICLR 2020 (11.5%) are on Label-efficient and robust learning, compared with 11.2% at ICLR 2019. They fall into 9 topics.
See ICLR 2020 in the interactive map
Submitting to ICLR? 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 · 31 papers
- Learning The Difference That Makes A Difference With Counterfactually-Augmented Data (229 citations)
- Decoupling Representation and Classifier for Long-Tailed Recognition (218 citations)
- Incorporating BERT into Neural Machine Translation (170 citations)
- Ensemble Distribution Distillation (124 citations)
- Learning Robust Representations via Multi-View Information Bottleneck (78 citations)
- Pitfalls of In-Domain Uncertainty Estimation and Ensembling in Deep Learning (73 citations)
- Weakly Supervised Disentanglement with Guarantees (67 citations)
- Rethinking the Hyperparameters for Fine-tuning (66 citations)
- FSPool: Learning Set Representations with Featurewise Sort Pooling (54 citations)
- Generalization through Memorization: Nearest Neighbor Language Models (53 citations)
- Disentangling Factors of Variations Using Few Labels (49 citations)
- Understanding and Improving Information Transfer in Multi-Task Learning (44 citations)
and 19 more in the interactive map.
Self-supervised and contrastive · 10 papers
- vq-wav2vec: Self-Supervised Learning of Discrete Speech Representations (415 citations)
- On Mutual Information Maximization for Representation Learning (274 citations)
- A Mutual Information Maximization Perspective of Language Representation Learning (108 citations)
- A critical analysis of self-supervision, or what we can learn from a single image (105 citations)
- Self-labelling via simultaneous clustering and representation learning (97 citations)
- Semantically-Guided Representation Learning for Self-Supervised Monocular Depth (47 citations)
- The Early Phase of Neural Network Training (47 citations)
- Adversarially Robust Representations with Smooth Encoders (12 citations)
- Self-Supervised Learning of Appliance Usage (6 citations)
- Target-Embedding Autoencoders for Supervised Representation Learning (4 citations)
Zero-/few-shot and meta-learning · 10 papers
- Multilingual Alignment of Contextual Word Representations (151 citations)
- Few-shot Text Classification with Distributional Signatures (121 citations)
- Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation (114 citations)
- Meta-Learning without Memorization (102 citations)
- Rapid Learning or Feature Reuse? Towards Understanding the Effectiveness of MAML (94 citations)
- Meta-Dataset: A Dataset of Datasets for Learning to Learn from Few Examples (74 citations)
- A Baseline for Few-Shot Image Classification (22 citations)
- Locality and Compositionality in Zero-Shot Learning (20 citations)
- Meta Dropout: Learning to Perturb Latent Features for Generalization (20 citations)
- A Theoretical Analysis of the Number of Shots in Few-Shot Learning (12 citations)
Semi-supervised learning · 7 papers
- DivideMix: Learning with Noisy Labels as Semi-supervised Learning (524 citations)
- ReMixMatch: Semi-Supervised Learning with Distribution Matching and Augmentation Anchoring (326 citations)
- Revisiting Self-Training for Neural Sequence Generation (139 citations)
- Automatically Discovering and Learning New Visual Categories with Ranking Statistics (34 citations)
- Unsupervised Clustering using Pseudo-semi-supervised Learning (16 citations)
- Semi-Supervised Generative Modeling for Controllable Speech Synthesis (15 citations)
- Difference-Seeking Generative Adversarial Network--Unseen Sample Generation (2 citations)
OOD and anomaly detection · 6 papers
- Classification-Based Anomaly Detection for General Data (94 citations)
- Your classifier is secretly an energy based model and you should treat it like one (81 citations)
- Deep Semi-Supervised Anomaly Detection (29 citations)
- Robust Subspace Recovery Layer for Unsupervised Anomaly Detection (25 citations)
- Novelty Detection Via Blurring (20 citations)
- Towards neural networks that provably know when they don't know (9 citations)
Noisy labels · 6 papers
- Curriculum Loss: Robust Learning and Generalization against Label Corruption (112 citations)
- SELF: Learning to Filter Noisy Labels with Self-Ensembling (54 citations)
- Learning from Rules Generalizing Labeled Exemplars (22 citations)
- Simple and Effective Regularization Methods for Training on Noisily Labeled Data with Generalization Guarantee (15 citations)
- Robust training with ensemble consensus (8 citations)
- Discrepancy Ratio: Evaluating Model Performance When Even Experts Disagree on the Truth (5 citations)
Domain adaptation and generalization · 4 papers
- Mutual Mean-Teaching: Pseudo Label Refinery for Unsupervised Domain Adaptation on Person Re-identification (407 citations)
- Federated Adversarial Domain Adaptation (146 citations)
- Domain Adaptive Multibranch Networks (15 citations)
- Masked Based Unsupervised Content Transfer (1 citation)
Test-time adaptation · 3 papers
Knowledge distillation · 2 papers
- Contrastive Representation Distillation (63 citations)
- Knowledge Consistency between Neural Networks and Beyond (20 citations)
