Institutional development · Digital transformation, data and AI · United States
Researchers demonstrate three new techniques to shield AI models from backdoor attacks
North Carolina State University reports that researchers have shown three techniques designed to stop attacks that make AI models behave maliciously or undesirably.
Why it matters: AI researchers working on model security have three newly demonstrated techniques, reported by North Carolina State University, aimed at preventing backdoor attacks.
North Carolina State University reported that researchers have demonstrated three new techniques intended to prevent so-called backdoor attacks on AI models.
According to the university, backdoor attacks are attacks that cause AI models to behave in malicious or undesirable ways.
The university describes the three techniques as a comprehensive suite for protecting AI.
Editor’s view
Sarah Williamson, Teaching, Digital and AI Editor
For researchers and developers concerned with AI model security, the result as reported offers three techniques aimed at preventing backdoor attacks that induce malicious or undesirable model behaviour.
.jpg){kind=link}