User-Centred AI/ML for an Automatic Time Use Diary for Digital Health
Project ID: 2228bd1224 (You will need this ID for your application)
Research Theme: Information and Communication Technologies
UCL Lead department: Division of Psychology and Language Sciences (PALS)
Lead Supervisor: Enrico Costanza
Project Summary:
Artificial Intelligence (AI) and Machine Learning (ML) provide great performance across an ever- increasing number of applications. However, due to the technical complexity of these systems, the large number of parameters and vast amounts of data involved in training them, a well-recognized challenge, known as human-centred AI, is to make them intelligible and usable by people who may not be technical experts in AI/ML. The overarching goal of this project is to advance human-centred AI, through the development of novel user interfaces and interaction techniques that leverage and extend existing AI explanations (e.g., LRP saliency maps). The PhD will provide training in AI/ML and human-computer interaction. The work will involve the iterative development of robust software prototypes and their experimental evaluation. It will involve a combination of technical work in AI/ML and user-centred design methods, working with people. Ideal candidates should have a strong technical background and interest in ML methods and human-computer interaction (and ideally experience with one of the two). To enable the project to follow an empirical approach, the work will include the development of a novel interactive software to classify images automatically captured by a wearable camera. Such software will leverage deep learning for image classification (one of the areas where deep learning is most successful) and interactive ML to overcome the limited availability of training data. This focus is motivated by use cases in digital health: to assist patients self-monitoring and self-caring for long-term conditions (e.g., diabetes), or affected by memory loss. As the nature of such a system is potentially privacy invasive, it will also provide opportunities to research ethical risks associated to ML. Most importantly, the software will act as a test bed to develop and rigorously evaluate novel interaction techniques around AI/ML, including explanation techniques.