#21 · ML / AI methods

Deep learning decodes EEG markers of consciousness, language coherence and prediction

The dominant thread is using learned classifiers and coherence measures to decode structured information from EEG in human cognition. A convolutional network classifies EEG responses to multi-dimensional transcranial electrical stimulation for consciousness measurement, reaching 92% holdout F1. A separate study argues that coherence in language unfolds across separable temporal regimes, with drift reflecting the gradual accumulation of meaning across extended context. A third framework decodes EEG-motion coherence for language prediction in Deaf signers, identifying left-hemispheric low-frequency features. What is new is the pairing of decoding with explicit coherence structure across stimulation, temporal language dynamics, and visually mediated prediction. These items are heterogeneous in scope and do not share a common latent-representation method, so the shared thread is methodological rather than mechanistic.