Computer Vision and Robotics
DESCRIPTION
Our computer vision and robotics research advances the state-of-the-art in deep learning architectures through both supervised and self-supervised learning paradigms. We develop efficient few-shot learning algorithms and continual learning frameworks that enable robust generalization from limited annotated data while supporting incremental adaptation to novel tasks and domains. Our work in higher-order deep learning frameworks tackles the challenges of heterogeneous data representation and fusion, developing tensor-based architectures that can effectively process and integrate multimodal information streams while preserving their inherent geometric and topological structures. Complementing these advances, we pioneer neuromorphic computing approaches that reimagine visual processing through event-based sensing and spiking neural networks. This bio- inspired architecture enables efficient computation of dynamic visual features, including optical flow, depth estimation, and visual odometry, while significantly reducing computational overhead compared to traditional frame-based deep learning systems. These theoretical and algorithmic advances drive innovations across industrial machine vision, medical image analysis, autonomous navigation, and robotic perception, where our frameworks demonstrate superior sample efficiency and adaptive capabilities in non-stationary environments.