AI ResearchMachine Learning1 min reading time

Adversarial Training Can Further Hide a Backdoor, but Does It Improve Trigger Identification?

LessWrong
Read the full article
Researchers examined whether adversarial training can better conceal backdoors in AI models and if it aids in identifying the triggers of such backdoors. The study explores the effectiveness of adversarial training in both hiding and detecting backdoor triggers.

More in AI Research