Finding Circuits in Protein Language Models

Description: Mechanistic Interpretability is a sub field of Natural Language processing which seeks to reverse engineer the behavior of black box models such as Transformers. The goal of this talk is to introduce key ideas from this field and discuss their applications to current Protein Language models like ESM-2.



 

Date:
Tuesday, March 17, 2026
Start Time:
12:10 p.m. PDT
Host or Sponsor:
Location:
Hybrid

M3-A808 and  Zoom

Speaker or Presenter:
Contact Information: