Robust Pre-Training by Adversarial Contrastive Learning - CAP6412 Spring 2021

Explore robust pre-training techniques through adversarial contrastive learning in this 44-minute lecture from the University of Central Florida. Delve into key concepts including contrastive learning, unsupervised adversarial training, and the Dual Stream algorithm. Examine experimental results, testing accuracy, and semisupervised adversarial training methods. Gain insights from an ablation study and robustness dynamics analysis before reaching the concluding remarks. Enhance your understanding of advanced machine learning strategies for improved model resilience and performance.

Intro
Background
Motivation
Contrastive Learning
Recap
Unsupervised adversarial training
Standard Simclair
Methodology
Dual Stream
Dual Stream Algorithm
KL Divergence
Semisupervised training
Experimental results
Testing accuracy
Semisupervised adversarial training
Ablation study
Robustness dynamics
Conclusion