Kevin Lu
lu.kev@northeastern.edu
Bio
I'm Kevin, a machine learning intern at Takeda Pharmaceuticals (training SAEs for antibody protein design) and an undergraduate researcher in the Bau Lab. I'm working on mechanistic interpretability of text-to-image diffusion models, protein language models, and protein structure generators. I'm interested in CS PhD programs focused on interpretable generative modeling and accelerating drug discovery pipelines.
Check out this interactive "DecoderLens" for ESMFold demo! -->
How does ESM-2 build up useful structural information across layers?
DecoderLens ESM language model encoder using folding trunk
News
- Nov 2025 — Served as a reviewer for ICLR Main Conference (Geometric Deep Learning).
- Sep 2025 — Our paper When are concepts erased from diffusion models? accepted to NeurIPS Main Track!
- Aug 2025 — Served as a reviewer for NeurIPS Mechanistic Interpretability Workshop.
- Aug 2025 — Presented To Bind or Not to Bind: A Layer-Wise Dissection of Binding Information in ESM at New England Mech Interp (NEMI) Workshop.
- Jul 2025 — Started co-op at Takeda Pharmaceuticals: focusing on protein language model interpretability for antibody protein design.
- Jun 2025 — Presented Where do concepts go in diffusion models? CVPR Generative Models for Computer Vision Workshop (Kaiming He was the main speaker!) and Visual Concepts Workshop!