About me

I'm Roy Segal, an ML researcher working on brain signals, video and physical systems.

I start with the data and the problem: which representations are useful, how to evaluate a model, and what it takes to make it work in practice. My work spans multimodal learning, computer vision and scientific forecasting.

Publications

arXiv preprint · 2026

STST-JEPA

A self-supervised transformer that learns EEG representations by predicting masked regions in latent space. Pretrained on 47,703 recording sessions, it transfers to age prediction and other EEG tasks with lightweight task-specific training.

Read the paper ↗

My works

Current · brain.space ↗

EEG foundation models

I'm currently developing foundation models for EEG at brain.space, learning representations of brain activity that can support different downstream tasks.