Qunjie Huang

Representation Learning and Generalization for Visual Brain Decoding

I am a Ph.D. Candidate in Computer Science at Yunnan University, studying how visual information is represented, transferred, aligned, and reliably decoded from human brain signals.

My research centers on learning transferable neural representations across individuals, connecting brain activity with visual-semantic spaces, and understanding how learned representations support reliable inference when neural evidence is limited or shifts across subjects.

Research themes

Neural Representation Learning. I study how subject variation shapes neural representations and how EEG encoders can generalize to previously unseen individuals.

Brain–Vision–Language Alignment and Distillation. I investigate how brain signals can be aligned with shared visual-semantic spaces and how knowledge can be transferred across modalities.

Reliable Neural Inference. I examine when learned representations remain reliable under subject shift, limited evidence, and deployment constraints.

My recent work, SATTC, accepted to CVPR 2026, is a case study in reliable neural inference: it uses retrieval geometry and label-free calibration to improve cross-subject EEG-to-image retrieval without retraining the encoder.

My current work primarily focuses on EEG, with broader interests in MEG, fMRI, ECoG, and algorithm–system co-design for future neurotechnology.