← davidlitman.com

AI / Neuro

Salk Institute — computational neuroscience

Courtyard and reflecting pool at the Salk Institute, La Jolla
Salk Institute, La Jolla

I currently work in the computational neuroscience lab of Francis Crick Chair Professor Terry Sejnowski at the Salk Institute.

Right now I am investigating the internal ontologies of transformer-based chess engines. My current focus is Maia-3, which takes Elo (the standard metric for chess skill) as an input to the pre-trained network, so you can vary the skill the network is conditioned on with no change to its weights. Turning that Elo dial up from 700 to 2500:

  1. Pushes the computation deeper, monotonically, for every chess piece and move type I measured.
  2. This "depth migration" happens most for specific tactics, especially knight forks.
  3. The mechanism appears to consist of deeper (later) heads getting recruited for more specialized computations while shallow (earlier) heads keep a roughly constant contribution.

How Does Changing the Elo of a Chess Transformer Affect its Computations?

Attention weights of layer 7, head 12 at Elo 2400: the position on the left, and an 8 by 8 grid of boards on the right showing where each square attends

ABCD dataset work

I spent a year running clustering on behavioral and imaging data from the ABCD study to identify a participant subgroup and investigate neural and genetic (GWAS) correlates. Publication upcoming as soon as final data set is released for research from NIH.

UCSB LEAP lab — emotional temporal coding

Contributed to a graduate student’s fMRI dissertation on emotional temporal coding, running multivoxel pattern analysis (MVPA) and representational similarity analysis (RSA) with region-of-interest masks in frontal pole and related areas (FSLeyes).

Recent books

Relevant to neuroscience

Reading list

Favorite papers, and a rough sketch of my deepest interests.