SpecMoE: Spectral Mixture-of-Experts Foundation Model for Cross-Species EEG Decoding

2026-03-17

AI Summary

This work addresses the limitation of existing EEG foundation models, which rely on single-domain (time- or frequency-only) masking during self-supervised pretraining and consequently exhibit bias toward high-frequency components while failing to adequately capture low-frequency neural rhythms—hindering cross-subject and cross-species generalization. To overcome this, we propose SpecHi-Net, a novel framework that integrates a joint time–frequency–spectrogram masking strategy via Gaussian-smoothed masks applied to short-time Fourier transform representations. Combined with a U-shaped hierarchical architecture and a spectral-gated Mixture-of-Experts (SpecMoE) mechanism, SpecHi-Net enables balanced representation learning across both high- and low-frequency neural activities. The method substantially improves EEG reconstruction fidelity and cross-species generalization, achieving state-of-the-art performance across diverse tasks—including sleep staging, emotion recognition, motor imagery, anomaly detection, and drug effect prediction—with consistently superior results on both human and mouse EEG datasets.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Computer Vision: Multi-modal Vision

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Decoding the orchestration of neural activity in electroencephalography (EEG) signals is a central challenge in bridging neuroscience with artificial intelligence. Foundation models have made strides in generalized EEG decoding, yet many existing frameworks primarily relying on separate temporal and spectral masking of raw signals during self-supervised pretraining. Such strategies often tend to bias learning toward high-frequency oscillations, as low-frequency rhythmic patterns can be easily inferred from the unmasked signal. We introduce a foundation model that utilizes a novel Gaussian-smoothed masking scheme applied to short-time Fourier transform (STFT) maps. By jointly applying time, frequency, and time-frequency Gaussian masks, we make the reconstruction task much more challenging, forcing the model to learn intricate neural patterns across both high- and low-frequency domains. To effectively recover signals under this aggressive masking strategy, we design SpecHi-Net, a U-shaped hierarchical architecture with multiple encoding and decoding stages. To accelerate large-scale pretraining, we partition the data into three subsets, each used to train an independent expert model. We then combine these models through SpecMoE, a mixture of experts framework guided by a learned spectral gating mechanism. SpecMoE achieves state-of-the-art performance across a diverse set of EEG decoding tasks, including sleep staging, emotion recognition, motor imagery classification, abnormal signal detection, and drug effect prediction. Importantly, the model demonstrates strong cross-species and cross-subject generalization, maintaining high accuracy on both human and murine EEG datasets.

Problem

• EEG decoding
• cross-species generalization
• spectral bias
• foundation model
• low-frequency rhythms

Innovation

• Spectral Mixture-of-Experts
• Gaussian-smoothed masking
• STFT-based reconstruction
• Cross-species EEG decoding
• SpecHi-Net