Active learning for adaptive brain machine interface based on Software Agent

Brain Machine Interface (BMI) and Software Agent (SA) can provide some new adaptive strategies for robust BMI implementations. In this work, a non-invasive Adaptive BMI is introduced, which has been designed to discriminate four mental tasks. The SA allows tracking features to contribute for an adaptive process, while the user´s engagement state provides a feedback between BMI and the environment. The Silhouette´s width is the performance measurement used for the active learning process. The results show that the implemented system allows high accuracy (75%) in the classification process.