UTSAV A. NEGI

I am a graduate researcher at Purdue University focusing on federated learning, reinforcement learning, and distributed optimization. My work bridges the gap between rigorous mathematical theory and scalable systems implementation, designing learning algorithms that converge reliably in resource-constrained, heterogeneous environments like wireless networks.

My research perspective is grounded in a strong foundation of distributed systems engineering. Early in my career, I built large-scale software architectures, an experience that revealed the critical performance challenges that emerge when theoretical models meet real-world constraints. During my Master's studies, I pivoted to research-driven inquiry, contributing to a project on parameter-free federated temporal difference learning (submitted to IEEE Transactions on Automatic Control). For this work, I led the experimental validation, re-architecting the simulation infrastructure in Julia to achieve a 30x performance speedup, which enabled robust testing across hundreds of heterogeneous agent configurations.

I operate on the conviction that theoretical guarantees must be validated at scale to drive real-world impact. I specialize in developing high-performance simulation frameworks that rigorously test mathematical bounds against empirical reality. My PhD objective is to deepen my expertise in stochastic approximation and convex optimization while building the systems infrastructure necessary to prove these ideas in practice. My research interests center on communication-efficient learning, multi-agent coordination, and trustworthy AI—specifically where convergence must be guaranteed under adversarial or resource-limited conditions.

Beyond research, I have served as a Graduate Teaching Assistant for Game Theory, mentored student developers in software best practices, and contributed to open-source formal verification tools. I am committed to a research culture that values not just novel algorithms, but also the reproducible infrastructure that makes scientific discovery verifiable and extensible.