Endowing the Machine with Active Inference: A Generic Framework to Implement Adaptive BCI

Recent developments in computational neuroscience gave rise to an efficient generic framework to implement both optimal perceptual (Bayesian) inference and choice behaviour. This framework named Active Inference rests on minimizing free energy or surprise. We suggest it could be used to implement efficient adaptive Brain-Computer Interfaces (BCIs). We briefly illustrate it on a simulated P300-speller task.