3

We introduce a Latent Clustering-based Selection method to choose a core subset from the entire unlabeled dataset, aiming to improve the efficiency of self-supervised adversarial training while preserving robustness.

We show how prior claims about black-box access sufficing for optimal membership inference do not hold for most useful settings such as SGD, and validate our findings with a new white-box inference attack.

We introduce the Weight-Curvature Index (WCI), a novel metric that captures the interplay between model parameters and loss landscape curvature to better understand and improve adversarially robust generalization in deep learning.

We propose AutoDefense, a response-filtering based multi-agent defense framework that filters harmful responses from LLMs.

Study how to certify and train for cost-sensitive robustness using randomized smoothing.

We propose an availability poisoning attack for generating transferable poisoned data across different victim learners.

We propose an extragradient-type method EDAC for generating less certain adversarial examples to improve robust generalization

Advocate to understand the concentration of measure phenomenon regarding inputs regions with high label uncertainty

A new nonconvex algorithm for robust phase retrieval, which attains a linear rate of convergence and improved statistical error rate.