Arman Hatami

Interested in learning and forgetting

[monosov_lab @MonosovLab]

Congrats Arman Hatami and Romina Aalishah!

[monosov_lab @MonosovLab]

The fact that visual bias is transferred among “AI” networks **without human intention** has deep relevance for application (eg, military decision making) and for thinking about **learning among/across biological social agents**!

Paper accepted at WACV.

https://t.co/mnWBjhMGs0

[Arman Hatami @Arman__Hatami]

Can a model transfer a hidden preference through random numbers?
Yes. We show subliminal learning occurs in vision-language models. Preferences and even wording pass from teacher to student via unrelated data when doing knowledge distillation.
WACV 2027
https://t.co/Tvamr5qTEd

[monosov_lab @MonosovLab]

Our posters this SFN!!! Come see us and say hi! (img by fbl2.0)