Channel Adaptation for EEG Foundation Models: A Systematic Benchmark Across Architectures, Tasks, and Training Regimes

2026-04-24

AI Summary

This study addresses the challenge of scaling EEG foundation models across heterogeneous electrode layouts by establishing the first systematic benchmark for channel adaptation that spans diverse model architectures, downstream tasks, and training paradigms. It evaluates four adaptation strategies: Conv1d projection, spherical spline interpolation, source-space decomposition, and Riemannian recentering. The findings reveal that the optimal adaptation method is highly architecture-dependent; while flexible architectures can intrinsically adapt during fine-tuning, they still require external adaptation when the encoder is frozen. Notably, negative transfer is observed between probing and supervised fine-tuning. Experiments demonstrate that CBraMod, with only 5M parameters, outperforms a general-purpose model 31 times larger in parameter count on four out of five datasets, underscoring the efficacy of compact, EEG-specific architectures.

Technology Category

• Cognitive Modeling & Cognitive Systems: Adaptive Behavior
• Machine Learning: Deep Neural Architectures and Foundation Models
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Practical large-scale studies of user experience
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Scaling EEG foundation models requires pooling data across heterogeneous electrode montages, a prerequisite both for larger pretraining corpora and for downstream deployment. We present the first systematic comparison of four channel adaptation methods (Conv1d projection, spherical spline interpolation (SSI), source-space decomposition, and Riemannian re-centering) across five pretrained EEG foundation models (5M--157M parameters), five downstream tasks, and two training regimes with 10--15 random seeds each. We find that rigid-montage models (BENDR, Neuro-GPT) require external adaptation, while flexible models (EEGPT, CBraMod) match or exceed it natively when fine-tuned but benefit from external methods under frozen-encoder deployment. A probe-SFT asymmetry exists: external adaptation can cause severe negative transfer during fine-tuning of flexible models. The optimal method is architecture-dependent (Conv1d for BENDR, SSI/Riemannian for Neuro-GPT, source-space decomposition for depression detection), and 5M-parameter CBraMod outperforms models up to 31$\times$ larger on 4/5 datasets, consistent with independent findings that compact EEG-specific architectures can match larger models.

Problem

• EEG foundation models
• channel adaptation
• electrode montages
• heterogeneous data
• model deployment

Innovation

• channel adaptation
• EEG foundation models
• systematic benchmark
• flexible vs rigid montage
• compact architecture