/reading-list

Paper-Reading

Paper reading list in representation learning, with special emphasis on self/semi-supervised learning, adversarial robustness, generalization ability, information theory and relevant topics.

Self-supervised Learning

Theory

  • [1]: "A Theoretical Analysis of Contrastive Unsupervised Representation Learning". ICML(2019) [PDF]
  • [2]: "THEORETICAL ANALYSIS OF SELF-TRAINING WITH DEEP NETWORKS ON UNLABELED DATA". ICLR(2021)) [PDF]

Contrastive Learning

  • [1]: "LEARNING WEAKLY-SUPERVISED CONTRASTIVE REPRESENTATIONS". ICLR(2022) [PDF]
  • [2]: "Weakly Supervised Contrastive Learning". ICCV(2021)[PDF]
  • [3]: "Representation Learning via Invariant Causal Mechanisms". ICLR(2021) [PDF]
  • [4]: "REPRESENTATION LEARNING VIA INVARIANT CAUSAL MECHANISMS". ICLR(2021) [PDF]
  • [5]: "Contrastive Learning Inverts the Data Generating Process". ICML(2021) [PDF]

Data Augmentation

  • [1]: "Estimating Training Data Influence by Tracing Gradient Descent". NeurIPS(2020) [PDF]
  • [2]: "Understanding the Generalization Benefit of Model Invariance from a Data Perspective". NeurIPS(2021) [PDF]

Information Theory and Its Application

  • [1]: "LEARNING WEAKLY-SUPERVISED CONTRASTIVE REPRESENTATIONS". ICLR(2022) [PDF]
  • [2]: "CONTRASTIVE REPRESENTATION DISTILLATION". ICLR(2020) [PDF]

Adversarial Learning

  • [1]: "CAT: Customized Adversarial Training for Improved Robustness". Arxiv(2020) [PDF]
  • [2]: "Perceptual Adversarial Robustness: Defense Against Unseen Threat Models". ICLR(2021) [PDF]

Semi-supervised Learning

  • [1]: "Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning". TPAMI(2017) [PDF]
  • [2]: "Time-Consistent Self-Supervision for Semi-Supervised Learning". ICML(2020) [PDF]
  • [3]: "FlexMatch: Boosting Semi-Supervised Learning with Curriculum Pseudo Labeling". NeurIPS(2021) [PDF]