Ari Benjamin

I study how and why neural networks learn what they do. I see neural networks both as models of learning in the brain and as intelligent systems in their own right, and I have spent the past decade training them and moving between these two views.

Much of my work is theory. I have shown that efficient neural codes emerge naturally from gradient descent (Nature Communications, 2022), studied how to measure and regularize networks in the space of functions they compute (ICLR, 2019), and recently showed that a network can be understood as a Bayesian ensemble of its tangent functions — a lens that reframes catastrophic forgetting and points toward new algorithms for continual learning (NeurIPS, 2024).

I also like to build. I designed and trained TissueFormer, a transformer that extends single-cell foundation models to predict tissue-level phenotypes from populations of single cells (BMC Bioinformatics, 2026), and have worked on uncertainty-aware objectives for post-training language models.

The brain is still my favorite learning machine. I am especially drawn to what neural networks don’t yet capture: neuromodulators like serotonin, the wild diversity of cell types, and how brains keep learning for a lifetime without overwriting what they know. Lately this has me thinking about belief formation and the self-consistency of knowledge in large language models — what is learned in context versus in weights, and what complementary learning systems might look like in modern AI. These strike me as questions that could be understood analytically, and I’d love to see that happen.

Along the way I have co-organized a COSYNE workshop on why networks learn what they do, co-authored lecture notes on connectionist theory with Andrew Saxe, Jay McClelland, and colleagues at the Gatsby Unit’s Analytical Connectionism summer school, and reviewed for NeurIPS, ICML, and Nature Communications.

I am currently a postdoctoral fellow in the laboratory of Tony Zador at Cold Spring Harbor Laboratory. I completed my PhD with Konrad Kording at the University of Pennsylvania.