Welcome! I’m Petros, a 4th year Ph.D. student in Operations Research at MIT, advised by Prof. Rahul Mazumder. My interests lie at the intersection of mathematical optimization, high-dimensional statistics, and machine learning.

In the summer of 2026, I interned at LinkedIn’s AI Foundations team within CoreAI, conducting research on Mixture-of-Experts (MoE) language model compression.

I was previously a Senior Research Analyst in the Banking Studies department of the NY Fed Research & Statistics Division, working on numerous research and policy projects regarding bank reserves and their effect on monetary policy implementation. I received my B.S. degree in Applied Mathematics (with a concentration in Computer Science) and Economics from Yale, graduating with honors and distinction in both majors.

Aside from problem solving, my hobbies include geography, calisthenics, swimming, biking, rowing, and listening to rap & contemporary classical music. I hold Greek and US dual citizenship.

email: pprastak AT mit DOT edu

Publications

(* denotes equal contribution)

• Differentially Private High-dimensional Variable Selection via Integer Programming
  Petros Prastakos, Kayhan Behdin, Rahul Mazumder
  NeurIPS 2025
  Preliminary version appeared at the ICLR 2024 Workshop on Private ML
  Presented at TPDP 2026 and 2026 Columbia University Workshop on Robust Statistics

• Reconstruction-Optimized Expert Pruning for Mixture-of-Experts Language Models
  Jelena Markovic-Voronov*, Petros Prastakos*, Kayhan Behdin, Jincheng Cao, Zhipeng Wang, Yuanda Xu, Zhengze Zhou, Wenhui Zhu, Shayan Mohajer Hamidi, Rohit K Patra, Rahul Mazumder
  Preprint; under review at ICLR 2027

• Interpretable Multistudy Learning with Similarity-Coupled Decision Rules
  Petros Prastakos, Gabriel Loewinger, Brian Liu, Rahul Mazumder
  Preprint