New EEG-Language Model Aligns Brain Signals with Semantics.

Researchers have developed the Brain Latent Predictive Model (BLPM), an innovative EEG-language foundation model designed to overcome limitations in current EEG decoding paradigms. Instead of focusing on low-level signal reconstruction or discrete token spaces, BLPM reframes heterogeneous EEG decoding tasks as a continuous semantic embedding prediction problem. This approach aims to better align continuous neural dynamics with natural language semantics. BLPM incorporates a Continuous EEG Latent Predictive (CELP) encoder, which learns transferable representations through latent target prediction. Complementing this, a Multi-Query Semantic Decomposition (MQSD) module extracts task-relevant information and aligns these continuous EEG representations with textual semantics within a shared latent space, based on their semantic relationships. Experiments across various benchmarks demonstrate BLPM's consistent generalization performance across diverse tasks, establishing continuous latent semantic prediction as an effective new paradigm for EEG-language foundation models.