Research

Our lab develops practical AI technologies that can learn from limited, imperfect, and real-world data. Our research focuses on computer vision, machine learning, and multimodal AI, with applications in medical AI, surveillance, scientific discovery, and trustworthy AI systems.

Research Overview

Weakly/Semi-supervised Learning

We study methods that reduce annotation costs by learning from weak, noisy, or partially labeled data, especially for visual recognition and segmentation.

Active Learning and Data-efficient AI

We develop active learning methods that select informative samples under limited labeling budgets, with a focus on low-budget and real-world annotation scenarios.

Domain Adaptation and Generalization

We aim to build robust models that maintain performance under distribution shifts across domains, environments, sensors, and institutions.

Multimodal AI

We explore AI systems that combine vision, language, audio, and other modalities for richer real-world understanding.

AI + X

We apply AI to interdisciplinary domains such as healthcare, science, engineering, and real-world systems, with an emphasis on reliability, interpretability, and practical impact.