Dr. Prasan Yapa is a Research Scientist and Assistant Professor at the University of Luxembourg and a Specially Appointed Researcher at Kyoto University of Advanced Science. He also serves as a Visiting Lecturer at Informatics Institute of Technology, teaching undergraduate and postgraduate computer science programs. He earned his Doctor of Engineering degree from Kyoto University of Advanced Science, Japan, specializing in ubiquitous and personal computing. His research interests include Natural Language Processing, Machine Learning, Deep Learning, Agentic AI, Digital Health Informatics, and Data Science. Dr. Yapa has extensive academic and industry experience, having worked as a Research Officer at A*STAR Institute for Infocomm Research, Tech Lead at Virtusa, and Senior Software Engineer at IFS. He is an active member of several professional organizations, including IEEE, Association for Computing Machinery, the International Neural Network Society, and the Information Processing Society of Japan.                            

Graph attention encoders for digital mental health applications in multi-party conversations   Depression detection from online text is an emerging area in digital mental health research. Multi-party conversations present unique challenges due to complex discourse structures and dynamic interactions among multiple participants. In this talk, we explore a graph attention-based encoder designed to model conversational dynamics for digital mental health applications. The approach introduces hierarchical representations by incorporating both root- and sub-level utterances, enabling deeper contextual understanding of dialogue structure. It also leverages multiple edge types to capture discourse relationships between utterances, speaker-specific characteristics, and temporal dependencies throughout the conversation. This design facilitates the representation of interpersonal interactions, the interpretation of individual behavioural cues, and the tracking of emotional shifts over time. To support comprehensive mental health assessment, a multi-task learning framework is employed to jointly model depression detection and depression severity classification. The approach is evaluated across several related tasks, including depressed utterance detection, depressed interlocutor recognition, and depression severity screening. Experimental findings demonstrate strong performance across these tasks, highlighting the potential of graph attention-based conversational modelling for advancing automated analysis of multi-party interactions in digital mental health contexts