Subhro Das

I am a Staff Research Scientist at the MIT-IBM AI Lab in IBM Research. As a Research Affiliate at MIT and Principal Investigator (PI), I work on developing novel deep learning algorithms in collaboration with MIT researchers.

My research interests are broadly in the areas of Representation Learning, Generative AI, Foundation Models, Trustworthy Machine Learning, Large Language Models, Reinforcement Learning, Dynamical Systems and ML Optimization methods. At the MIT-IBM AI Lab, my research work focuses on long-context understanding, uncertainty quantification and human-centric system design for Large Language Models, and time-series foundation models.

Currently, working on post-training techniques and alignment methods for LLMs, especially in fine-tuning, instruction tuning, synthetic data generation and prompt optimization. Generating large-scale long-context instruction following data, with variety of reasoning tasks, for training Granite models. Contributing in the post-training stage of multimodal model developments, primarily in the Speech-Language and Vision-Language models.

My research papers are published in top machine learning and signal processing venues, including ICML, NeurIPS, ICLR, AAAI, AISTATS, EACL, Transactions on Machine Learning Research, IEEE Transactions on Signal Processing and ICASSP. I am an IBM Master Inventor and have filed 25+ patents in machine learning.

Education

PhD, Electrical & Computer Engineering, Carnegie Mellon University, 2016. Advised by: Prof. José Moura.

MS, Electrical & Computer Engineering, Carnegie Mellon University, 2013.

B.Tech., Electronics & Communication Engineering, Indian Institute of Technology Kharagpur, 2011.

Appointments

IBM Research
2018 - present: Staff Research Scientist / Research Manager, AI Models. MIT-IBM AI Lab, Cambridge, MA.
2016 - 2018: Research Scientist, ML for Health. IBM T.J. Watson Research Center, Yorktown Heights, NY.

Massachusetts Institute of Technology
2019 - present: Research Affiliate, Electrical Engineering & Computer Science. Cambridge, MA.

Current MIT-IBM Research Grants

• Human-Centric AI: Novel Algorithms for Shared Decision Making
  PI: David Sontag (MIT), Subhro Das (IBM), Dennis Wei (IBM), Prasanna Sattigeri (IBM)
• Principles and Methods for Exploiting Unlabeled Data in Supervised Learning
  PI: Greg Wornell (MIT), Subhro Das (IBM), Prasanna Sattigeri (IBM)
• Safe Learning for Time Series Problems: Data, Structure and Optimization
  PI: Luca Daniel (MIT), Subhro Das (IBM), Lam Nguyen (IBM)
• Adaptive, Robust, and Collaborative Optimization
  PI: Ali Jadbabaie (MIT), Asu Ozdaglar (MIT), Subhro Das (IBM), Nima Dehnamy (IBM), Songtao Lu (IBM)
• Coarse Graining Using Machine Learning
  PI: Tommi Jaakkola (MIT), Nima Dehmamy (IBM), Subhro Das (IBM)

Students

Mentored/supervised some outstanding graduate students during their internship and scholar programs at IBM Research: Quang Nguyen (PhD, MIT), Tuomas Oikarinen (PhD, UC San Diego), Maohao Shen (PhD, MIT), Eli Lucherini (PhD, Princeton), Kadeem Noray (PhD, Harvard), Ran Xin (PhD, CMU), Nicholas Borge (MS, MIT), Yingying Li (PhD, Harvard), Joshua Lee (PhD, MIT), Renzhe Yu (PhD, UC Irvine), Orlando Romero (PhD, RPI), Nathan Hunt (PhD, MIT). Aside from them, collaborated with several students and postdocs from multiple universities.