Research focus

01   Learning from brain signals at scale

We explore self-supervised objectives that learn from the structure of brain data itself, including reconstructing missing information and relating different views of an observation. We ask how spatial, temporal, and anatomical information should shape pretraining for fMRI and EEG.

02   Shared structure across modalities

Brain recordings and the experiences that evoke them offer complementary perspectives. We investigate ways to connect neural representations with visual, auditory, language, and behavioral information, while preserving modality-specific properties and accommodating incomplete observations.

03   Transfer that respects individual variation

A useful representation should support adaptation beyond its training setting. We study cross-subject and cross-task transfer, efficient adaptation, and robustness to changes in recording conditions. We ask what should be shared across people and what should remain personalized.

How we investigate

We connect pretraining, adaptation, and downstream evaluation within a common research workflow. Directions include fMRI-based decoding and reconstruction, EEG task decoding, and multimodal representation learning. We assess transfer on held-out people or conditions and examine robustness, data efficiency, and interpretability alongside predictive accuracy.

Across our research framework

Brain foundation models turn insights from neural representation research into reusable computational tools. They also provide a testing ground for brain-inspired architectures and a potential basis for decoding in closed-loop NeuroAI systems.

fMRI & EEG · Self-supervised learning · Multimodal representations · Cross-subject transfer