Research Interests

My research interests lie broadly in artificial intelligence for visual understanding, with a focus on methods that are robust, interpretable, and applicable to real-world problems.

Computer Vision

I am interested in how machines perceive, segment, localize, and reason about visual information. This includes both classical image understanding tasks and emerging research directions in temporal and multimodal vision.

Medical Imaging

A major part of my research focuses on chest X-ray analysis, disease localization, and lung segmentation. I am particularly interested in building clinically relevant methods that go beyond raw accuracy and better address trust, uncertainty, and explanation.

Explainable AI

I explore techniques that help interpret the reasoning of deep learning models, especially in applications where transparency matters. My work examines visual explanation methods such as saliency and activation mapping, and how they can be evaluated more rigorously.

Weakly Supervised Learning

I am interested in developing methods that learn from limited or coarse annotations, especially in medical imaging where expert labels are expensive and difficult to obtain.

Uncertainty-Aware AI

Reliable deployment requires models to know when they may be wrong. I am interested in uncertainty estimation, failure analysis, and confidence-aware systems for segmentation and diagnosis pipelines.

Video Matting and Temporal Vision

I am interested in temporal consistency, edge fidelity, and memory mechanisms for video understanding problems, particularly in tasks such as real-time background matting.

Research Statement

My research aims to build AI systems that are not only accurate, but also interpretable, robust, and usable in real-world settings. I am especially interested in domains where trust matters, such as healthcare. Across my projects, I try to combine principled methodology, strong evaluation, and practical deployment thinking so that the resulting work contributes both to science and to real applications.