Facial Expression Interpretation in ASD Using Deep Learning

People with autism spectrum disorder (ASD) are known to show difficulties in the interpretation of human conversational facial expressions. With the recent advent of artificial intelligence, and more specifically, deep learning techniques, new possibilities arise in this context to support people with autism in the recognition of such expressions. This work aims at developing a deep neural network model capable of recognizing conversational facial expressions which are prone to misinterpretation in ASD. To that end, a publicly available dataset of conversational facial expressions is used to train various CNN-LSTM architectures. Training results are promising; however, the model shows limited generalization. Therefore, better conversational facial expressions datasets are required before attempting to build a full-fledged ASD-oriented support system.

Automated summary

This research aims to develop a deep neural network model capable of recognizing conversational facial expressions which are prone to misinterpretation in individuals with autism spectrum disorder. Promising training results were achieved, but the model showed limited generalization, highlighting the need for better datasets before building a full-fledged support system for ASD.