Researcher
ML & AI sparse neural networks, structured connectivity, neural network initialization, pruning, fine-tuning, LLM pretraining, lottery ticket hypothesis.
Math arithmetic combinatorics, additive number theory, approximate groups, algebraic graph theory, expander graphs.
Latest
Machine learning and AI
Why Sparse Structure Helps a Transformer Learn Its Second Skill
A small arithmetic transformer can master addition yet stall when multiplication is introduced. Biregular sparse initialization changes which connections survive and, at extreme sparsity, makes full multiplication generalization substantially more likely.
Graph theory
Cops, Robbers, and Algebraic Graphs
Algebraic structure turns pursuit on Cayley and Cayley sum graphs into a countdown, yielding cop-number bounds in terms of the generating set and progress toward the Meyniel conjecture.