Hello!
I’m Rareș, I am a graduating PhD candidate at the Operations Research Center at MIT, and I am delighted to be advised by Professor Georgia Perakis. I completed my undergrad in Computer Science at the Georgia Institute of Technology, finishing my thesis with Santosh Vempala.
Research Focus
At the core of my research lies the exploration of novel algorithms and optimization techniques to address the intricate challenges that emerge in today’s interconnected world at the intersection of machine learning and optimization. My work aims to bridge the gap between theoretical advancements and practical implementations to improve decision-making.
I have worked in many aspects of end-to-end learning, an area of learning machine learning models whose predictions are fed into downstream optimization tasks. I have answered questions such as how to train models in order to minimize downstream decision cost? how to train models to make decisions robust against uncertainty and worst-case scenarios? and how can we learn causality when decisions lie in an overarching optimization problem? Overall, how can AI benefit OM applications?
I have collaborated with IBM and Boston Scientific on applications of these ideas and their impact in inventory management and pricing. I have also gained additional practical experience through internships at Oracle and Meta.
Featured Projects
ProjectNet: Fast End-to-End Learning
The true measure of a learning model’s quality lies in the tangible impact of the decisions it influences, particularly in the context of downstream optimization/decision-making tasks.
We introduce ProjectNet, a method to computationally-efficiently integrate the prediction and optimization stages together.
Accepted AAAI 2023
Expanded version revised for Management Science
Discretization for Robust Optimization
Robustness to uncertainty is a crucial property in real-world decision-making which builds trust when faced with high-risk scenarios.
We provide a novel framework based on discretizing the feasible region aimed at minimizing expected cost as well as being robust and protect against worst
case scenarios.
Accepted ICLR 2024
Expanded version revised for Management Science
Learning under Decision-Dependent Uncertainty
The decisions one takes can often affect the outcome observed. How can we effectively learn to make decisions when there are no ground-truth counterfactual observations?
We tackle this problem by constructing uncertainty sets over the space of ML models and present efficient algorithms to optimize over these sets.
Submitted ICLR 2025
Expanded version soon to be submitted to Operations Research