Reproduce “Beyond $L_\infty$-Norm Sets: Synthesizing Interpretable Non-Uniform Uncertainty Sets for Evaluation-Time State-Adversarial Robustness” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xulin Chen, Ruipeng Liu, Garrett E. Katz
PD-NUUS sets state- and coordinate-specific perturbation bounds for evaluating pretrained continuous-control policies. It ties coordinate-wise observation bounds to a tolerance on expected discounted-return loss, using surrogate constraints and approximations. On HalfCheetah with PPO, the maximum degradation among four tested attackers was 34.2, versus 399.2 under an L-infinity radius of 0.20. The method can overshoot its target; it is not a guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.