Adversarial Contrastive
Adversarial contrastive learning combines contrastive learning's self-supervised approach to representation learning with adversarial training's focus on robustness. Current research emphasizes applications across diverse domains, including image classification, spatial-temporal graph learning, and natural language processing, often employing variations of transformer-based models and adapting the technique for federated learning scenarios. This approach aims to improve the quality and robustness of learned representations, particularly in situations with limited labeled data or noisy/biased datasets, leading to more accurate and fairer models for various applications.
Papers
August 16, 2024
June 19, 2023
July 22, 2022
June 7, 2022
May 26, 2022