Prateek Chanda

I am a 4th year Ph.D. student at the Indian Institute of Technology Bombay, CSE Department, advised by Pratik Jawanpuria and Ganesh Ramakrishnan, working on large-scale efficient machine learning, generously supported by the Prime Minister Research Fellowship (2022–2027) and the Microsoft PhD Award (2025–2026).

I have been a Student Researcher (twice) at Google DeepMind India, hosted by Pradeep Shenoy during (July 2023–Nov 2023 & Dec 2023–May 2024), working on Efficient Pretraining of Large Language Models and Knowledge Distillation for Large Language Models for proprietary Gemini and LaMDA models. During my time at DeepMind, I had the pleasure of working with and being mentored by Dheeraj Nagaraj, Arun Suggala and Prateek Jain.

Recently, I also interned at Microsoft Research India, where I worked on generalizable and efficient test-time scaling with Nagarajan Natarajan, Amit Sharma and Subbarao Kambhampati (see our work on Interwhen).

The central theme of my research is building efficient and scalable machine learning systems. My work spans efficient LLM pretraining and fine-tuning (1), continual learning (2), and submodular methods for data selection (3). Going forward, I am broadly interested in the following research directions for improving the capabilities and reliability of large models.

Efficient LLM pretraining and fine-tuning. I am interested in data selection objectives and training strategies that improve sample efficiency for large language models, especially under data and compute constraints. A direction I find especially promising is minibatch selection via submodular methods and gradient matching for better pretraining.
Continual and federated learning. I am interested in methods for continual learning that enable models to adapt over time without catastrophic forgetting, and in federated approaches for personalized learning. I am especially interested in constrained fine-tuning methods that preserve previously learned knowledge.
Knowledge distillation and model compression. I am interested in robust distillation techniques that improve the efficiency and reliability of smaller models trained from large teacher models, including robustness to early-layer representations.

Before my Ph.D., I was a Research Software Engineer (SCAI Fellow) at Microsoft Research India Bangalore, hosted by Amit Sharma and Vageesh Chandramouli. Before that, I completed my undergrad at IIEST Shibpur, advised by Prof. Susanta Chakrabarty (former Dean) and Prof. Asit Kr. Das.

Fun Fact: During my undergrad, I was an astrophysics enthusiast and often contemplated career choices. I spent much of that time contributing to SunPy (a Solar Physics Python library) alongside researchers at NASA GSFC, and in 2024 got recognized with the NASA Group Achievement Award.

Select Publications & Preprints

Minibatch Selection for Language Models via Partition Matroid Constrained Gradient Matching Prayas Agrawal*, Prateek Chanda*, Ishita Khatri, Ganesh Ramakrishnan, Bamdev Mishra, Pratik Jawanpuria
ICML 2026 [arxiv] [code] [talk] [presentation] [poster]
UniPROT: UNIform PRototype Selection via Partial Optimal Transport with Submodular Guarantees Prateek Chanda*, Prayas Agrawal*, Karthik S. Gurumoorthy, Ganesh Ramakrishnan, Bamdev Mishra, Pratik Jawanpuria
AISTATS 2026 [arxiv] [code] [presentation] [poster]
Interwhen: A Generalizable Framework for Verifiable Reasoning with Test-time Monitors Vishak K Bhat*, Prateek Chanda*, Vijval Ekbote*, Ashmit Khandelwal, Maitreyi Swaroop, Vineeth N. Balasubramanian, Subbarao Kambhampati, Nagarajan Natarajan, Amit Sharma
Preprint · ICLR 2026 Workshop LLM Reasoning · ICLR 2026 Workshop VerifAI-2 [arxiv] [code] [Microsoft Talk]
Sculpting Subspaces: Constrained Full Fine-Tuning in LLMs for Continual Learning Nikhil Shivakumar Nayak, Krishnateja Killamsetty, Ligong Han, Abhishek Bhandwaldar, Prateek Chanda, Kai Xu, Hao Wang, Aldo Pareja, Oleg Silkin, Mustafa Eyceoz, Akash Srivastava
ICLR 2026 [arxiv] [code] [talk] [presentation] [poster]
FairPO: Fair Preference Optimization for Multi-Label Learning Soumen Kumar Mondal, Prateek Chanda, Akshit Varmora, Ganesh Ramakrishnan
NeurIPS OPT Workshop 2025 [pdf] [code] [talk] [presentation] [poster]
Bandit Guided Submodular Curriculum for Adaptive Subset Selection Prateek Chanda*, Prayas Agrawal*, Lokesh Reddy Polu, Saral Sureka, Atharv Kshirsagar, Ganesh Ramakrishnan
NeurIPS 2025 [openreview] [code] [talk] [presentation] [poster]
Few Shot Chain-of-Thought Driven Reasoning to Prompt LLMs for Open Ended Medical Question Answering Saeel Sandeep Nachane*, Ojas Gramopadhye*, Prateek Chanda*, Ganesh Ramakrishnan, Kshitij Sharad Jadhav, Yatin Nandwani, Dinesh Raghu, Sachindra Joshi
EMNLP 2024 Findings [paper] [code] [talk] [presentation] [poster]
Bayesian Coreset Optimization for Personalized Federated Learning Prateek Chanda, Shrey Modi, Ganesh Ramakrishnan
ICLR 2024 [openreview] [code] [talk] [presentation] [poster]
Would I Regret Later Joining This Community? Using Temporal Neighborhood Information for Community Retention in a Game Theoretic Community Detection Framework Prateek Chanda, Susanta Chakrabarty
AAAI-22 Workshop MLOR [arxiv] [code] [talk] [presentation] [poster]
Robust Deep Reinforcement Learning Control Framework Prateek Chanda
AAAI-22 Workshop on Robust Artificial Intelligence System Assurance [arxiv] [code] [talk] [presentation] [poster]

* denotes equal contribution.

Select Awards & Honours

AISTATS Financial Award, 2026
Microsoft PhD Award, 2025–2026
Qualcomm Innovation Fellowship Finalist, 2025 and 2026
ICLR Travel Grant, 2024
Prime Minister Research Fellowship (PMRF), 2022–2027

News & Coverage

Our proposal on Efficient Federated Finetuning for Language Models was accepted to Project Tapestry (Chief Scientific Advisor: Yann LeCun). [Christopher Nguyễn's tweet] [IIT Bombay's tweet]

Yann LeCun reposted our work. [LinkedIn]

Service

Teaching

2026 Spring	CS 218: Design and Analysis of Algorithms (Sujoy Bhore)
2025 Autumn	PMRF TA-ship: Subset Selection for Large Language Models CS 419: Introduction to Machine Learning (Ganesh Ramakrishnan)
2025 Spring	CS 769: Optimization in Machine Learning (Ganesh Ramakrishnan)
2024 Autumn	PMRF TA-ship: Convex Optimization & Submodular Methods CS 602: Applied Algorithms (Sujoy Bhore)
2024 Spring	CS 769: Optimization in Machine Learning (Ganesh Ramakrishnan)
2023 Autumn	PMRF TA-ship: Advanced Topics in Machine Learning CS 105: Discrete Structures (S. Akshay)
2023 Spring	CS 419M: Introduction to Machine Learning
2022 Autumn	CS 335: Artificial Intelligence and Machine Learning; CS 337: AI and ML Lab

Software

Efficiency Learning Research project on efficient ML pretraining and learning.

Interwhen Generalizable framework for verifiable reasoning with test-time monitors (Microsoft Research).

Decile Data-efficient machine learning toolkit.