Task Specific Signal Transformation

Neural signals, sourced directly from the brain, are the gold standard for use in cognitive state detection, but are infeasible for everyday usage. However, the use of more accessible non-neural data to predict cognitive state is significantly less accurate. To bridge this gap in accuracy and usability, in this paper we propose a method that uses non-neural data to infer neural activity and improve off-body cognitive state detection. This is achieved by exploiting the inherent correspondence of signals collected simultaneously from a single individual, and learning a task-specific signal transformation from non-neural to neural data. This reconstructed neural data representation can then be used to predict cognitive state with 87% accuracy, recovering 94% of the accuracy obtained using neural signals directly. This represents a significant advance in practical human–machine-teaming for everyday usage, allowing neural information to be inferred from non-neural sources. In other words, the neural information becomes tacit.