Academic work
I am interested in the theory and practice of statistical learning, machine learning, and optimization, with a broader curiosity about inference, uncertainty, and decision-making in complex systems. Here are some projects that I have worked on.
I led a four-person project studying lightweight, causal test-time adaptation under distribution shifts. We compared normalization-based strategies on synthetic time-series benchmarks and financial data, finding when adaptation helps—and when conservative batch-statistics updates can be more reliable.
We model how viewers, human creators, and AI-generated content respond to platform recommendations and crowd-out effects. Using equilibrium analysis, spectral stability conditions, and simulations, we explore human-creator dropout and the tension between short-run welfare and long-run ecosystem health.
We study nonparametric multivariate prediction with temporally dependent covariates, relating the approximation capacity of sparse recurrent and feedforward networks to statistical risk. The project develops vector-output risk bounds under mixing assumptions and examines recurrent and feedforward architectures in simulations.
Working manuscript in preparation.I compared evaluation benchmarks for data-analysis agents and investigated a framework spanning statistical correctness, insight quality, visualizations, decision relevance, and reproducibility. I also developed Python experiments and analytical outputs to support evaluation on financial and retail data.