Double Seminar - Statistics

Speakers: Saverio Ranciati (Department of Statistical Sciences, University of Bologna) and Hongyng Meng (Brunel)

12.00: Hongyng Meng (Brunel)

Title: Emotion Recognition from Facial Expression Videos and its Applications

Abstract:

Facial expression analysis has becoming a popular research topic in recent years due to multidiscipline collective efforts from researchers in computer science, psychology, and cognitive science. Artificial intelligence has made significant contribution for facial expression analysis that can be used for the design of advanced human machine interaction system, intelligent robots and computer games. It can also be used for human mental health analysis such as dementia, autism and clinical diagnosis application such as shoulder pain and low back pain. This talk will address the problem on how to capture emotion information from facial expressions by generate statistic and dynamic features and how to use advanced machine learning methods to modelling the dynamic of the facial expression sequences to achieve the higher recognition rate. The talk will cover the 2 international challenging winning works as well as some recent work based deep learning. In the end, the applications of the facial expression recognition will be viewed.

13.00:  Saverio Ranciati (Department of Statistical Sciences, University of Bologna)

Title: "Bayesian finite mixture model with overlapping clusters for the analysis of network data"

Abstract:

Network data usually summarize information about the relationships between individuals and their interactions as communities. Their social behaviours are used to understand which features tie them together, providing an insight about the group structure of the whole network by detecting the link connecting each unit. In some cases, these intertwined dynamics are characterized by records of individuals (actors) attending to events. A common approach is to project actor-events data into an actor-actor setting, which sometimes provides biased and contaminated answers. We focus instead on a model-based clustering approach that encodes the natural actor-events representation of the data, allowing for observations to be assigned to (potentially) more than one community via a Bayesian finite mixture model with multiple allocations. The driving research question is to understand the group structure of some real world dataset: i.e., a terrorist network dataset, where recordings of militants and their attendance to different meetings and operations are available.

Buffet lunch at 12.45.