ELLIS Talk by Hayley Hung

In search of the optimal model for social conversational interaction.

Abstract

In this talk, I will describe a problem I have been working on in the last decade to estimate and model conversing groups in crowded social networking events. Like much of my research, I combine social science theory with machine learning to solve machine perception tasks. I will describe the nature of this problem that may initially seem quite simple and hopefully highlight to you the complexities of digging deeper into modelling human social behaviour. Concretely, I will be discussing the role of graph clustering and graph neural networks in solving this problem, as well as the pitfalls of evaluation metrics, and data labelling and collection.

To join this event, please contact Frans Oliehoek.