EvoBrain: Continual Learning of EEG Foundation Models Across Heterogeneous BCI Tasks

Researchers introduce EvoBrain, a continual learning framework that enables EEG foundation models to adapt across multiple brain-computer interface tasks without catastrophic forgetting. The system uses neural-spectral normalization and distillation techniques to balance learning new tasks while retaining knowledge from previous ones, advancing toward unified brain decoding systems.

EvoBrain represents a significant technical advancement in brain-computer interface research by solving a fundamental scalability problem in EEG foundation models. Traditional approaches require separate fine-tuning for each task, creating computational overhead and preventing knowledge transfer across different brain-signal interpretation tasks. This research introduces a continual learning framework that allows a single model to sequentially learn multiple BCI tasks while maintaining performance on earlier ones—a challenge known as the plasticity-stability tradeoff.

The breakthrough stems from two key innovations: Neuro-Spectral Task Normalization handles the unique characteristics of brain signals by normalizing new tasks against historical data patterns, while Response-Affinity Distillation preserves the model's learned representations from previous tasks and enables selective knowledge sharing between compatible task types. This architectural approach directly addresses limitations in current foundation model deployment, where linear scaling of resources with task count creates practical barriers.

For the brain-computer interface industry, this work reduces deployment costs and accelerates development cycles. Organizations building multi-application BCI systems can now implement genuinely universal models rather than maintaining task-specific deployments. The framework's validation across six distinct BCI tasks demonstrates robustness, suggesting practical applicability across different brain-signal applications from medical monitoring to assistive technologies.

The research positions continual learning as essential for next-generation brain-computer systems. Future development will likely focus on scaling these methods to larger task volumes and integrating real-time adaptation capabilities, potentially enabling adaptive BCIs that improve through clinical use.

• →EvoBrain enables EEG foundation models to learn multiple BCI tasks sequentially without forgetting previous knowledge
• →The framework reduces computational overhead that scales linearly with task count in conventional multi-task approaches
• →Neuro-Spectral Task Normalization and Response-Affinity Distillation address the plasticity-stability tradeoff in brain-signal processing
• →Evaluation across six BCI tasks shows consistent performance improvements over existing state-of-the-art methods
• →The work advances toward unified, one-for-all brain decoding systems applicable across diverse clinical and assistive applications