Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster B in Poster Session B: Tuesday, August 4, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Back to the Feature: Toward a Feature-Centric Account of Brain–LM Alignment

Timna Wharton Kleinman¹, Ariel Goldstein¹; ¹Hebrew University of Jerusalem

Presenter: Timna Wharton Kleinman

Large Language Models (LLMs) have emerged as powerful proxies for linguistic processing in the human brain, yet the standard practice of quantifying this alignment via a single scalar obscures the underlying drivers of the relationship. This "scalar-centric" approach treats high-dimensional embeddings as atomic units, establishing that alignment exists while remaining agnostic to how it is achieved. We propose a shift toward a feature-centric perspective that inspects the specific embedding dimensions contributing to neural alignment. We demonstrate that while naive feature analyses suggest a superficial homogeneity across the brain, this is a methodological artifact. By synergizing feature decomposition via Sparse Autoencoders (SAEs) with sparse encoding models (Lasso), we uncover distinct feature subsets that dissociate between cortical regions and temporal windows. We redefine brain-LLM alignment not as aggregate similarity, but as a structural inquiry into which computational dimensions map onto distinct neural dynamics.

Topic Area: Methods, Tools, Theory & Neural Coding