research
Research on generalizable neural representations, brain–vision–language alignment, and reliable visual brain decoding.
I study visual brain decoding as a connected learning problem: first learning neural representations that transfer across individuals, then aligning them with visual-semantic knowledge, and finally making inference reliable when neural evidence is limited or shifted.
Research themes
1. Neural Representation Learning
How can we learn visual neural representations that preserve meaningful information while generalizing across people?
This theme focuses on cross-subject EEG encoders, the structure of subject variation, and representation-learning principles for previously unseen individuals.
2. Brain–Vision–Language Alignment and Distillation
How can neural activity be connected to the structured knowledge already captured by vision and language models?
This theme studies shared visual-semantic spaces, cross-modal alignment, and knowledge distillation across brain-signal and computational modalities.
3. Reliable Neural Inference
How can learned neural representations remain reliable under subject shift, limited evidence, and deployment constraints?
This theme studies the reliability boundaries of visual brain decoding. SATTC is one concrete case study, using retrieval geometry and label-free calibration to examine inference under cross-subject distribution shift.
