Sort by
Refine Your Search
-
Category
-
Program
-
Employer
-
Field
-
representations remain confidential throughout the collaborative workflow. Defenses against model extraction and chain-of-thought extraction attacks, designing protocols that detect and mitigate adversarial
-
to ensure that model updates, gradients, and intermediate representations remain confidential throughout the collaborative workflow. Defenses against model extraction and chain-of-thought extraction attacks
-
date may also apply. In that case, documentation from the principal supervisor regarding the expected date of thesis submission and defence must be provided. A possible appointment is then conditional
-
getting these defenses to run on small, low-power embedded devices, so the lab develops lightweight AI and trustworthy embedded platforms that can detect and respond to threats on their own. The goal is
-
to cyber attacks that can cause physical damage. AI:HARDWARE uses AI to protect this infrastructure and to design hardware that is secure by design. A key challenge is getting these defenses to run on small