Social Signal and Multimodal Processing

Alessandro Vinciarelli

University of Glasgow, UK

Alessandro Vinciarelli is Full Professor at the University of Glasgow (School of Computing Science and Institute of Neuroscience and Psychology),where he works since 2010. Previously, he has been with several companies in Italy and the USA (Finmeccanica,Accenture and IBM) and with the Idiap Research Institute in Switzerland. His main research interest is Social Signal Processing, the domain aimed at modelling, analysis and synthesis of nonverbal behaviour in social interactions. He has published more than 140 works and he is or has been PI and co-PI of more than 15 national and international projects, including the European Network of Excellence on Social Signal Processing ( He has chaired more than 25 international events - including the IEEE International Conference on Social Computing and the ACM International Conference on Multimodal Interaction - and he is a co-founder of Klewel (, a knowledge management company recognized with several national and international awards.

Course Description

The course will revolve around the development of multimodal approaches for the analysis of human behaviour. In particular, the course will show how the concept of multimodal communication, originally formulated in ethology and cognitive sciences, has been articulated in Artificial Intelligence terms. In the first part, the course will show how the information extracted from multiple sensors (the modalities) can be modelled jointly through the adoption of two main methodologies:

  • Early fusion: the information extracted from multiple modalities is combined before statistical modelling through the creation of a multimodal feature vector, i.e., a vector where every component is a physical measurement extracted from one of the multiple input streams
  • Late fusion: the information extracted from every individual modality is modelled statistically independently of the others and, then, the outcome of the various statistical models is combined.

In the second part, the course will provide an introduction to recent approaches made possible by the advent of deep neural networks.


November 30th, 2018