Discovering Mental Strategies for Voluntary Control Over Brain-Computer Interfaces.

When novices engage with a brain-computer interface, the first challenge is to discover what mental strategy allows some rudimentary control over the system. In some instances, the strategy is clear and provided by the trainer. In most cases, the strategy is not always known. In neurofeedback training, the goal is to increase or decrease a particular brain signature, whilst participants discover which mental strategies are helpful. The question is how? The starting point is the assumption that reinforcement learning underlies the neurofeedback learning and is involved in evaluating the utility of a candidate strategy. Candidate strategies are proposed via activation of neural representations of task goals that are part of a neural network that includes the target brain region. Early in the learning stage, positive feedback allows the accumulation of activation in this neural network until the task goal representation (i.e., the mental strategy) becomes fully activated. Low levels of reward prediction error lead to the candidate strategy to remain activated. High levels of reward prediction error lead to a specific deselection of the candidate strategy and allows selection of a new candidate. The process of selection and deselection within this mental space demonstrates area-restricted search, which is an optimal search strategy in situations where rewards are heterogeneously distributed, typically clustered, over a certain area. Further investigation into strategy discovery is needed to advance our current understanding of neurofeedback learning, its relation to reinforcement learning, and the role of conscious awareness.