PhD defense Tristan Cladière: "Context-aware emotion recognition: exploring compact architectures for social robotics""Context-aware emotion recognition: exploring compact architectures for social robotics"

"Context-aware emotion recognition: exploring compact architectures for social robotics"

Abstract:

In order to improve human-robot exchanges, it is necessary to exploit both verbal and non-verbal information emitted by humans, and to integrate them into the decision-making process of robots. The main non-verbal indicator of a person's emotional state is its facial expression, but other clues can also be considered: the posture, the action perform, the location represented, or even the visible objects. All these elements provide context, which helps to extrapolate the emotion. In the literature, this is called Context-Aware Emotion Recognition (CAER). It is in this research framework that we have developed an innovative bottom-up approach, called BENet, generating its predictions in the form of emotion maps. In addition to multitask learning, we have also explored knowledge amalgamation to merge the knowledge of several models (teachers) within BENet (student). Finally, we focused on Visual Emotion Analysis (VEA), and introduced a novel multi-teacher knowledge distillation method, allowing a simple neural network to outperform much more complex approaches in the literature.

Jury:

• Alice CAPLIER (Pr) - GIPSA-lab, Ecole Phelma Grenoble - Rapporteure
• Mohamed DAOUDI (Pr) - CRIStAL, IMT Nord Europe Lille - Rapporteur
• Alice OTHMANI (MdC HDR) - LISSI, Université Paris-Est - Examinatrice
• Fabrice LEFEVRE (Pr) - LIA, Avignon Université - Examinateur
• Olivier ALATA (Pr) - LabHC, Université Jean Monnet Saint-Etienne, Directeur de thèse
• Christophe DUCOTTET (Pr) - LabHC, Université Jean Monnet Saint-Etienne, Co-directeur de thèse
• Hubert KONIK (MdC) - LabHC, Université Jean Monnet Saint-Etienne, Co-encadrant
• Anne-Claire LEGRAND (MdC) - LabHC, Université Jean Monnet Saint-Etienne, Co-encadrante