"Trustworthy AI-human interaction in cyber-physical systems", Leader CI, Australian Data61/CSIRO CRP Project grant C020996. Awarded $130,440. 2019-2022.

In this project, we aim to develop testing and validation technologies to enhance the trustworthiness of machine learning models trained on edge devices. As a new form of AI models training, we first will use new software development patterns to effectively integrate data from multiple modalities for training AI-Human Interaction models. As a new form of AI testing, we plan to adopt data-driven metamorphic relations to thoroughly test AI-Human Interaction models to detect models' safety issues. As a new form of runtime verification for AI models, we will develop an online-learning anomaly detection architecture that can detect anomalies in the AI-Human interaction models at runtime. As a case study, we will create a variety of AI models for driver fatigue detection using semi-autonomous or autonomous vehicles.