Unsupervised Domain Adaptation for Facial Emotion Recognition in Autistic Children

Autism is a neurodevelopmental disorder characterized by deficits in social, interpersonal interaction and communication skills. A generalized facial emotion recognition model does not scale well when confronted with the emotions of autistic children due to domain shift inherent distributions source (neurotypical) target (autistic) population. The dearth labeled datasets field autism exacerbates problem. Domain adaptation using generative adversarial (GAN) counters this disparity creating an that aligns features domains training. This paper looks at building classifier can identify idiosyncrasies associated child’s expression generating feature-invariant representations distribution. objective two-fold – a) build discriminative accurately b) train feature generator produce invariant representation taking into account their similar yet different data distributions, presence unlabeled data. Investigation automatic classification expressions population has been pursued extensively vis-a-vis neurotypical complexities eliciting interpreting obtained from children.