All the following papers can be found on iclr2022.
- Information-Theoretic Generalization Bounds for Iterative Semi-Supervised Learning.
- How unlabeled data improve generalization in self-training? A one-hidden-layer theoretical analysis.
- Improved Fine-tuning by Leveraging Pre-training Data: Theory and Practice.
- Towards Understanding the Data Dependency of Mixup-style Training.
mixup - How does Contrastive Pre-training Connect Disparate Domains?.
- Explaining Scaling Laws of Neural Network Generalization.
- DIVA: Dataset Derivative of a Learning Task.
trainable dataset - Towards Understanding Data Values: Empirical Results on Synthetic Data.
data valuation - Data-centric Semi-supervised Learning.
- Resolving Training Biases via Influence-based Data Relabeling.
influence function - Understanding the Success of Knowledge Distillation -- A Data Augmentation Perspective.
DA vs KD - Training Data Size Induced Double Descent For Denoising Neural Networks and the Role of Training Noise Level.
- Theoretical Analysis of Consistency Regularization with Limited Augmented Data.
data augmentation - Self-Supervise, Refine, Repeat: Improving Unsupervised Anomaly Detection.
- Less data is more: Selecting informative and diverse subsets with balancing constraints.
- Data-Dependent Randomized Smoothing.
- From SCAN to Real Data: Systematic Generalization via Meaningful Learning.
- Adaptive Control Flow in Transformers Improves Systematic Generalization.
- Iterative Decoding for Compositional Generalization in Transformers.
- Reference-Limited Compositional Learning: A Realistic Assessment for Human-level Compositional Generalization.
- [Compositional Attention: Disentangling Search and Retrieval](Compositional Attention: Disentangling Search and Retrieval).
- Icy: A benchmark for measuring compositional inductive bias of emergent communication models.
- Illiterate DALLE Learns to Compose.