Scaling Next-Brain-Token Prediction for MEG

TL;DR

This paper introduces a scalable autoregressive model for MEG data that can generate long sequences of brain activity, demonstrating stability and specificity across multiple large datasets.

Contribution

It presents a novel approach combining a vector-quantizer and a large backbone model to generate and evaluate long-horizon MEG sequences across datasets.

Findings

Generates stable long sequences of MEG data

Achieves cross-dataset generalization

Outperforms control conditions in specificity tests

Abstract

We present a large autoregressive model for source-space MEG that scales next-token prediction to long context across datasets and scanners: handling a corpus of over 500 hours and thousands of sessions across the three largest MEG datasets. A modified SEANet-style vector-quantizer reduces multichannel MEG into a flattened token stream on which we train a Qwen2.5-VL backbone from scratch to predict the next brain token and to recursively generate minutes of MEG from up to a minute of context. To evaluate long-horizon generation, we introduce task-matched tests: (i) on-manifold stability via generated-only drift compared to the time-resolved distribution of real sliding windows, and (ii) conditional specificity via correct context versus prompt-swap controls using a neurophysiologically grounded metric set. We train on CamCAN and Omega and run all analyses on held-out MOUS, establishing cross-dataset generalization. Across metrics, generations remain relatively stable over long rollouts and are closer to the correct continuation than swapped controls. Code available at: https://github.com/ricsinaruto/brain-gen.