InPhyRe Discovers: Large Multimodal Models Struggle in Inductive Physical Reasoning, ArXiv, 2025
Incorporating Interventional Independence Improves Robustness against Interventional Distribution Shift, ArXiv, 2025
CoInD: Enabling Logical Compositions in Diffusion Models, ICLR, 2025
OASIS Uncovers: High-Quality T2I Models, Same Old Stereotypes, ICLR, 2025 (Spotlight)
Physics Informed Neural Network for Dynamic Stress Prediction
Research Experience
Worked as a software engineer at Qualcomm India.
Education
Ph.D. candidate supervised by Dr. Vishnu Naresh Boddeti at Michigan State University; B.Tech. in Electrical Engineering from Indian Institute of Technology Madras, undergraduate thesis advised by Dr. A.N. Rajagopalan.
Background
Research focuses on building safe and robust machine learning models. Applications include improving robustness against distributional shifts, evaluating inductive physical reasoning in LLMs, enhancing the controllability of diffusion models for compositional generation, and assessing stereotypes in T2I models.