1,057 followers
A new open-source toolset enables more precise evaluation and improvement of AI text classifiers by identifying single-word changes that can alter classification outcomes. doi.org/g9w27q
So far, Altmetric has seen
3 Bluesky posts from 2 Bluesky users, with an upper bound of 1,693 followers.
636 followers
MIT researchers developed techniques to improve text classifiers' resilience against single-word adversarial attacks. Their SP-Attack and SP-Defense methods enhance classifier robustness while cutting down on adversarial example generation costs. https:…
636 followers
Researchers devised a method to enhance text classifiers' security through single-word perturbations, exposing model vulnerabilities. SP-Attack and SP-Defense improved attack rates and classifier robustness, marking a significant step in AI application …