Affective Modeling for Children with Autism Spectrum Disorders: Application of Active Learning, pp. 297-310
Authors: Karla Conn Welch, Medha Sarkar, Nilanjan Sarkar, Changchun Liu, Vanderbilt University, Nashville, Tennessee, and others
Abstract: This chapter presents an overview of our research on the investigation of an affect-sensitive system to be applied in future autism intervention. A physiology-based affect-inference and adaptation framework was proposed, which could endow the assistive intervention technology with the capability of detecting the affective states of a child with Autism Spectrum Disorders (ASD) and responding to them accordingly. Given the importance of affective cues in human-machine interaction and its significant role in autism intervention practice, this chapter marks an important step towards intelligent intervention systems that embody human-like functionality - affect recognition and adaptation. To account for the spectrum nature of autism and the differences of emotional expression, an individual-specific approach was employed for affective modeling. Two computer-based cognitive tasks were designed for eliciting target affective states considered important in autism intervention. An active learning technique is developed in order to reduce the labeling requirement for developing the individual affective models. The preliminary results demonstrate that such an affect-sensitive adaptive system could hold promise for computer-assisted autism intervention.