Bio
I am a PhD Candidate in Electrical and Computer Engineering(ECE) at Yale University, advised by Professor Prof. Dionysis Kalogerias.
Prior to Yale, I completed dual Bachelor of Science degrees in Computer Engineering and Pure Mathematics at the Sharif University of Technology. Before that, I ranked 9th nationwide in Iran’s National University Entrance Examination and earned a silver medal in the Iranian National Mathematical Olympiad.
Research Interests
My research is organized around four interconnected directions, all centered on understanding the dynamics of, and optimization over, probability measures. These themes connect ideas from stochastic optimization, statistical learning, convex and nonconvex duality, information geometry, and decision-making under uncertainty.
Distributionally Robust Optimization (DRO). I study learning and decision-making problems in which the data-generating distribution is uncertain or misspecified. My work focuses on ambiguity sets, risk measures, optimal transport, and dual reformulations that lead to robust and statistically meaningful optimization procedures.
Federated Learning. I investigate distributed learning systems where data remain decentralized across clients. My research focuses on partial participation, client sampling bias, aggregation under distributional mismatch, and optimal-transport-based methods for aligning the training dynamics with a desired global objective.
Constraint Optimization and Duality Theory. I work on constrained statistical learning beyond convex settings, especially problems involving fairness, safety, coverage, or risk constraints. A central theme is understanding when Lagrangian methods are theoretically justified through zero duality gap, dense hypothesis classes, and universal PAC learnability.
Dynamics of Sampling. I study how sampling mechanisms affect learning, optimization, and long-run statistical behavior. This includes non-uniform participation, stochastic approximation, distribution shift induced by sampling, and the interaction between sampling dynamics and the objective being optimized.