We develop learning-based methods for environment perception from millimeter-wave radar imaging, with an emphasis on physics-aware representation learning, scalable pre-training, and language-guided open-world understanding. Our goal is to turn rich radar spectrum measurements into robust, transferable, and queryable representations for object detection and scene understanding under challenging sensing conditions.

We develop LLM- and VLM-driven methods to make remote sensing image analysis more scalable, semantically rich, and capable of reasoning. Our work spans large-scale vision-language data construction and reasoning-driven pixel-level understanding, connecting open geospatial data, language models, and universal segmentation models.

We developed digital visual-motor systems and algorithms to autonomously evaluate developmental coordination disorder (DCD) and delve into associated pathologies.

We developed deep learning methods to automatically discover action segments from the untrimmed videos.

We are developing SDF-based AI models for computational electromagnetics.

We are developing AI models for computational electromagnetics.

We developed a prior-guided framework for RSVP-EEG decoding and found that, for this task, a simple MLP can match or even outperform Transformer-based models.

We proposed a brain-machine fusion approach to achieve the brain-in-the-loop modeling and brain-out-of-the-loop application.

Boosting brain-computer interface performance through cognitive training: a brain-centric approach

We developed deep learning methods for quantitative muscle atrophy evaluation using ultrasound images.