Doctoral Speaking Skills Talk - Aviv Bick
October 28, 2026 10:00AM—11:00AM
Location:
6501
-
Gates and Hillman Centers
Speaker:
AVIV BICK,
Ph.D. Student, Computer Science Department, Carnegie Mellon University
https://avivbick.github.io/
State-space models offer an efficient alternative to attention, but often struggle to recall information over long contexts. In this talk, I’ll explain how organizing recurrent state into independently updated memory slots allows information to persist, and introduce Raven, a model built around this approach. I’ll present results showing improved retrieval and recall, including generalization to sequences sixteen times longer than those used in training.
Presented in Partial Fulfillment of the CSD Speaking Skills Requirement
Contact
Matt Stewart