Toward Generalizable and Interpretable Vision-language Decoding from the Human Brain: A Survey

PhD Qualifying Examination


Title: "Toward Generalizable and Interpretable Vision-language Decoding from
the Human Brain: A Survey"

by

Mr. Yulong LIU


Abstract:

Unraveling the operational mechanisms of the human brain remains a paramount
objective shared by both neuroscientists and artificial intelligence
researchers. Among various neural systems, the human visual and language
systems hold pivotal positions in the fields of brain science and computer
science. While the visual cortex occupies a significant portion of the
cerebral cortex, decoding perceptual experiences also engages broader neural
networks involved in semantic processing and cognition. Deciphering these
systems offers dual benefits: it inspires the development of more efficient
and robust bio-inspired AI architectures while simultaneously accelerating
advancements in Brain-Computer Interface (BCI) technologies. Driven by recent
breakthroughs in deep learning and generative models, significant progress
has been made in decoding visual and semantic information from brain signals.
This survey provides a comprehensive review of the field, systematically
organizing the development trajectory across six key dimensions: task
definitions, data resources, primary challenges, training paradigms, model
architectures, and future trends. Furthermore, we identify three promising
research directions: (1) Enhancing cross-subject generalization in few-shot
scenarios by leveraging rich cross-subject data to pre-train universal brain
decoding foundation models, requiring only minimal fine-tuning to mitigate
data sparsity issues inherent to single subjects; (2) Achieving zero-shot
cross-subject generalization for foundation models, enabling deployment on
new users without additional training; and (3) Improving the interpretability
of decoding models regarding vision-language information processing, such as
visualizing the neural mechanisms underlying visual imagery and semantic
reconstruction. Finally, the survey concludes with a brief overview of the
author's recent exploratory work focused on enhancing generalization and
interpretability in brain decoding.


Date:                   Monday, 15 June 2026

Time:                   11:00am - 12:00noon

Venue:                  Room 5560
                        Lift 27/28

Committee Members:      Prof. Yike Guo (Supervisor)
                        Dr. Sirui Han (Chairperson, EMIA)
                        Dr. May Fung