Mimicking adaptation processes in the human brain with neural network retraining

Abstract

Human brain processes undergo cycles of adaptation in order to meet the requirements of novel conditions. In affective state recognition, brain processes tend to adapt to new subjects as well as environmental changes. By using adaptive neural network architectures and by collecting and analysing data from specific environments we present an effective approach in mimicking these processes and modelling the way the need for adaptation is detected as well as the actual adaptation. Video sequences of subjects displaying emotions are used as data for our classifier. Facial expressions and body gestures are used as system input and system output quality is monitored in order to identify when retraining is required. This architecture can be used as an automatic analyzer of human affective feedback in human computer interaction applications. © 2007 International Federation for Information Processing.