Customized Cognitive State Recognition Using Minimal User-Specific Data

Automatic cognitive state recognition is very important for military training, rehabilitation, soldier safety, and mission success. However, developing cognitive state recognition algorithms is highly challenging due to the difficulties in building a generic model whose parameters fit all subjects. Further, it is very expensive and/or time-consuming to acquire user-specific training examples that allow the algorithms to be tailored for each user. We propose a generic machine learning framework to minimize the data acquisition effort, by combining transfer learning and active class selection. Transfer learning improves the recognition performance by combining a small number of user-specific training examples with a large number of auxiliary training examples from other similar subjects. Active class selection optimally selects the classes to generate user-specific training examples on-the-fly. We illustrate the effectiveness of the proposed approach on task difficulty classification from neural and physiological signals.