#14 · ML / AI methods

Neural decoders tested under naturalistic stimuli and continuous or real-time recording

The dominant thread is evaluating neural decoders under naturalistic or continuous conditions rather than against a single constrained target. A continuous EEG preprint uses a compact DNN to detect transition-related potentials marking narrative context at film cuts in continuous recordings. A real-time fMRI preprint adapts the MindEye2 pipeline to RT-Cloud, decoding single-trial visual perception within seconds. A published study decodes continuous valence and arousal from combined gray and white matter iEEG across two tasks, with cross-task evaluation. What is new is the emphasis on realistic regimes over a single decoding target. Modality and goal differ across these items, so the shared element is evaluation emphasis rather than a shared method.