Rahul Mazumder

NTU Associate Professor
Operations Research and Statistics group
MIT Sloan School of Management
Operations Research Center
LIDS
MIT Center for Statistics and Data Science
Massachusetts Institute of Technology

I am the NTU Associate Professor of Operations Research and Statistics, in the Operations Research and Statistics group at MIT Sloan School of Management. I am affiliated with the Operations Research Center (ORC), Laboratory for Information & Decision Systems (LIDS), MIT Center for Statistics. Outside MIT, I am also an Academic Scholar at LinkedIn AI (Algorithms and Foundations Group).
Before joining MIT, I spent a couple of wonderful years as an Assistant Professor at Columbia University (Dept of Statistics), where I was also affiliated with the Data Science Institute. Prior to Columbia, I was a PostDoctoral Associate at MIT. I completed my PhD in Statistics from Stanford University under the supervision of Trevor Hastie . I completed my BStat, MStat from Indian Statistical Institute, Kolkata.

Service

• I am currently serving/have recently served as an Associate Editor/Action Editor for the Annals of Statistics, Bernoulli, Operations Research (Optimization; and Machine Learning and Data Science), and Journal of Machine Learning Research.
• I am also serving as the series editor (Algorithms) of Cambridge University Press (Institute of Mathematical Statistics, TextBooks and Monographs)
• I was the founding cluster chair of Machine Learning within INFORMS Optimization Society

Research

I am a statistician working at the interface of statistics and operations research (mathematical optimization). I study problems in computational/algorithmic statistics using tools from mathematical optimization (convex and discrete optimization). I am interested in the applications of statistics/OR tools in problems arising in industry (e.g, computational finance, insurance, recommender systems etc), government (US Census Bureau) and the sciences (biostatistics, biomedical sciences). In the past I have worked on applications of statistics in the social sciences, oceanography, and turbulence.

Please take a look at the publications tab.

Principal Fields of Interest

• Statistics, Machine Learning, Mathematical Optimization (Convex optimization, Mixed Integer optimization), Large Scale Optimization Algorithms.
• High dimensional statistics and sparsity, combinatorial statistical modeling & computation, nonparametric function estimation (e.g., shape constrained inference).
• Applications of the above in recommender systems, computational finance, computational biology & healthcare, survey research, insurance pricing, etc. Neural Network Compression/Pruning, conditional computing in neural networks.

Recent News

• Papers accepted to NeurIPS 2025. Two papers on LLM compression under structural constraints, led by students from my group: Mehdi Makni and Xiang Meng. Third paper on differential privacy for best-subset selection using integer programming (with Petros Prastakos and Kayhan Behdin). Congratulations to Mehdi, Xiang, Petros and Kayhan.
• Excited to be a part of a collaboration with LinkedIn AI on algorithmic approaches for large model compression --- paper accepted to EMNLP '25 (Oral, Industry Track). The paper documents an industry application/deployment that uses a structured pruning algorithm designed by my group (Meng, Behdin,..,Mazumder, ICML '24)
• Some algorithms on foundation model compression developed my members of my research group have been implemented and open-sourced by LinkedIn AI at Github. Specifically, this repo features algorithms: ALPS (Meng et al NeurIPS '24 on unstructured, semi-structured pruning of LLMs), OSSCAR (Meng et al ICML '24; structured pruning) and Quantease (Behdin et al; quantization). Grateful for the collaboration!
• Two papers appeared in KDD '25 (main conference track): (i) Differential privacy for LLM finetuning (Makni et al and collaborators from Google), (ii) Stable rule extraction for interpretable machine learning led by Brian Liu.
  Additionally, two workshop papers: (iii) Error detection in Image-Caption pairs (Afriat et al and collaborators from IBM), (iv) Improving LLMs for tabular data using rules (led by Brian Liu)
• Paper on Sparse PCA (joint with Kayhan Behdin) using integer programming under a statistical model was accepted to the Annals of Statistics.

Contact Information

MIT Sloan School of Management
Building E62-583
100 Main Street
Cambridge, MA 02142
Ph: 617-253-2652
Email: rahulmaz "at" mit "dot" edu

Support Staff: Please see Sloan Website