all InfoSec news
Adversarial Robustness Limits via Scaling-Law and Human-Alignment Studies
April 16, 2024, 4:11 a.m. | Brian R. Bartoldson, James Diffenderfer, Konstantinos Parasyris, Bhavya Kailkhura
cs.CR updates on arXiv.org arxiv.org
Abstract: This paper revisits the simple, long-studied, yet still unsolved problem of making image classifiers robust to imperceptible perturbations. Taking CIFAR10 as an example, SOTA clean accuracy is about $100$%, but SOTA robustness to $\ell_{\infty}$-norm bounded perturbations barely exceeds $70$%. To understand this gap, we analyze how model size, dataset size, and synthetic data quality affect robustness by developing the first scaling laws for adversarial training. Our scaling laws reveal inefficiencies in prior art and provide …
accuracy adversarial alignment arxiv cs.cr cs.cv cs.lg gap human image law making problem robustness scaling simple sota studies understand unsolved
More from arxiv.org / cs.CR updates on arXiv.org
Jobs in InfoSec / Cybersecurity
Social Engineer For Reverse Engineering Exploit Study
@ Independent study | Remote
Application Security Engineer - Remote Friendly
@ Unit21 | San Francisco,CA; New York City; Remote USA;
Cloud Security Specialist
@ AppsFlyer | Herzliya
Malware Analysis Engineer - Canberra, Australia
@ Apple | Canberra, Australian Capital Territory, Australia
Product CISO
@ Fortinet | Sunnyvale, CA, United States
Manager, Security Engineering
@ Thrive | United States - Remote