On the Loss Landscape of Adversarial Training: Identifying Challenges and How to Overcome Them
This addresses optimization difficulties in adversarial training for machine learning models, offering a practical improvement but is incremental as it builds on existing adversarial training methods.
The paper tackled the problem of adversarial training leading to unfavorable loss landscapes with increased curvature and scattered gradients, which impede optimization and cause sharp minima. They introduced a periodic adversarial scheduling (PAS) strategy that overcomes these challenges, yielding better results and reduced sensitivity to learning rate compared to vanilla adversarial training.
We analyze the influence of adversarial training on the loss landscape of machine learning models. To this end, we first provide analytical studies of the properties of adversarial loss functions under different adversarial budgets. We then demonstrate that the adversarial loss landscape is less favorable to optimization, due to increased curvature and more scattered gradients. Our conclusions are validated by numerical analyses, which show that training under large adversarial budgets impede the escape from suboptimal random initialization, cause non-vanishing gradients and make the model find sharper minima. Based on these observations, we show that a periodic adversarial scheduling (PAS) strategy can effectively overcome these challenges, yielding better results than vanilla adversarial training while being much less sensitive to the choice of learning rate.