What objective does SAM approximately optimize?

Question: Sharpness-Aware Minimization modifies ordinary empirical-risk minimization by approximately minimizing which objective? A) The expected loss after adding isotropic Gaussian noise only to the input features B) The maximum training loss within a bounded neighborhood of the current parameters C) The trace of the exact Hessian subject to zero training loss D) The average loss of independently initialized models lying on a low-loss path Correct: B Explanation: SAM seeks parameters whose surrounding neighborhood also has low loss. In practice, it approximates a min-max problem by first finding an adversarial parameter perturbation and then updating the original parameters using the gradient evaluated near that perturbed point. Topic: advanced ML / optimization / sharpness-aware minimization