Research

Working papers

Adaptive multi-agent driving with population play.

in preparation·with Eugene Vinitsky·EMERGE Lab·2026

We train an "ego" agent against a population of conditioning-aware learning opponents in a driving simulator, then test how well it generalizes to held-out adaptive-adversaries. We study how architecture, opponent diversity, and reward design affect whether the agent adapts within a trial, and design a replication protocol that separates within-configuration variability from cross-cell differences.
Finite-size scaling of Schelling-type segregation.

in preparation·2026

We perform a finite-size scaling study of the Schelling segregation model and its multi-radius generalizations. We measure critical exponents at four radii on lattices up to 2048 x 2048, verify universality across radii, and demonstrate hyperscaling through data collapse. The result establishes a noise floor for the order parameter and clarifies which earlier claims survive at large n.
Adaptive sensing driven by prediction error.

in preparation·with Amin Arbabian·Arbabian Lab, Stanford EE·2026

We develop two adaptive sensing strategies built on a learned forward model: each next measurement maximizes the expected reduction in predictive uncertainty, concentrating samples where the structure is least known. The strategy requires under one-fifth the queries of random or active-learning baselines, and we give analytical bounds linking sample budget to the intrinsic dimensionality of the signal.

Forthcoming.