Host institution: UMIL Duration 36 Months
The use of Deep Learning in Human Computer Interaction and human performance is an emerging topic of research. The subject of this ESR will deal with the data analysis and the use of Deep learning approach in this field in order to detect and predict safety critical events in workplaces characterized by high complexity and automation. The research will use Big Data coming from body-signals, the human-machine interaction dynamics and the environmental settings (e.g. industrial processes). The main scope is to look for those signals that can represent reliable predictors of potential issues during the task execution. In fact, the factors that lead to a human error/failure are many (personal, organizational, contextual, etc.) and several are still unknown. In this context, applying Deep Learning algorithms to stream data, especially when dealing with several heterogeneous interconnected data streams is an open challenge and for this training network the challenge is also to compare what results it can produce with the results coming from more white boxes approaches, such as a Bayesian Network models developed in ESR3. This work will leverage recent work on explainable AI, and attention maps, to analyze the factors that the neural models attend to when predicting safety critical scenarios.
Investigate and test deep learning techniques for event detection and prediction in complex settings
Use explainable AI techniques to understand the factors the models attend to during prediction and compare and contrast these results with the Bayesian models developed in ESR3
A number of high-impact publications
Planned secondment(s): The PhD student is going to be seconded on M12 for 6 months in ADIENT to work on LIVE LAB 3 and on M12 to IMR for 12 months to work on LIVE LAB 1 and support the data collection and analysis from the LIVE LAB1 case study
to access the application form click on this link: Application form