2026-08-03 00:00 Papers Foundations & Methods Translated from EN

Preprint: STEAM Hierarchical Transfer Framework for EEG Decoding

Summary A preprint presents STEAM, a hierarchical transfer framework for EEG decoding that pairs a dual-branch spatio-temporal encoder with a shared soft mixture-of-experts (SSMoE) module to reconcile general representation learning with paradigm-specific specialization in EEG foundation models, letting complementary representations exchange information through a compact set of soft slots. The researchers report that STEAM attained the best average rank at competitive inference cost across 7 downstream datasets and 14 evaluation settings, and that its hierarchical pretraining further lifted decoding accuracy without retraining from scratch. The study is a preprint and has not been peer-reviewed.

Why it matters A shared soft mixture-of-experts that balances general and paradigm-specific learning shows a path for EEG foundation models to specialize without losing generality — the tension that now defines that research direction.

An arXiv preprint presents STEAM, a hierarchical transfer framework for EEG decoding that reconciles general-purpose representation learning with paradigm-specific specialization in EEG foundation models, using a dual-branch spatio-temporal encoder with a shared soft mixture-of-experts (SSMoE) module that aligns the spatial and temporal branches.

The researchers report that across 7 downstream datasets and 14 evaluation settings, STEAM attained the best average rank among the compared methods at a competitive inference cost, and its hierarchical pre-training strategy further specializes the model to a target paradigm without retraining from scratch, yielding consistent gains in decoding accuracy. This study is a preprint and has not been peer reviewed.

Compiled by BCIwiki from public sources

Sources · 1

arxiv.org 2026-08-03

Decoding Algorithms

Read original ↗

Suggest a correction Revisions · none 中 / EN