Soumi Das

Soumi Das is a Researcher at the Max Planck Institute for Software Systems (MPI-SWS), Saarbrücken, advised by Professor Krishna Gummadi. Her current research focuses on understanding Large Language Models (LLMs), including knowledge estimation, privacy-utility trade-offs, and language learning. Previously, she conducted her PhD at IIT Kharagpur under Sourangshu Bhattacharya, focusing on data-centric AI algorithms for subset selection and data valuation in machine learning and computer vision.

Education:

• PhD in Computer Science & Engineering (2017–2023), IIT Kharagpur
• M.Sc. Computer Science (2016), Banaras Hindu University
• B.Sc. Computer Science (2014), St. Xavier’s College, Kolkata

Research Interests:

• Foundations of LLMs: Knowledge estimation, memorization dynamics, privacy
• Data-centric AI: Subset selection, data valuation, optimization
• Applications in autonomous systems and computer vision

Key Contributions:

• Developed algorithms like CheckSelect and VTruST for efficient data valuation and subset selection
• Worked on defense mechanisms against adversarial attacks in federated learning
• Contributions to data-efficient autonomous driving through TMCOSS