PHD POSITION
Supélec - IMS Research Group
2 rue Edouard Belin, 57070 Metz - FRANCE
Title:
Imitation Learning and Task Transfer for Natural Human-Computer Interaction
Description:
Imitation learning (also referred to as apprenticeship learning or learning from demonstration) is a research field aiming at building machines with the ability of performing tasks after having observed a finite number of demonstrations by an expert. Task transfer is similar to imitation learning although the machine and the expert are not supposed to have the same capabilities. Then, the machine has to catch the goal of the expert so as to reach it using any possible ways. Inverse reinforcement learning (IRL) provides a theoretical solution to both these problems. In the IRL paradigm, the expert is supposed to receive rewards while performing the task. Therefore, it is assumed that the expert acts so as to collect a maximum of rewards. Yet, these rewards are hidden to the machine while observing the expert. Instead of imitating the expert in a supervised manner, IRL proposes to discover the reward function optimized by the expert from demonstrations so as to learn in turn a behavior that maximizes the collected rewards. The learnt reward function is a compact representation of the task. Apprenticeship learning via inverse reinforcement learning is a promising research domain in the field of robotics but has not been investigated in the domain of human-machine interaction. Yet, this is an obvious application since to reach naturalness, a human-computer interface (HCI), and especially a multimodal interface, should exhibit a behavior that is close to the one of a human.
The selected candidate will investigate the theoretical foundations of IRL so as to develop efficient algorithms able to learn from a limited number of demonstrations. Then, these algorithms will be applied to build natural human-computer interfaces exhibiting an adaptive and social behavior.
Context:
This research is part of a starting project funded by the European Commission under the FET OPEN program. The ILHAIRE project (Incorporating Laughter into Human-Avatar Interactions: Research and Evaluation) is a 3-year project aiming at studying the role of laughter in human-human and human- machine interactions. The candidate will thus evolve in an international consortium regrouping partners from France, Belgium, Italy, Switzerland, United Kingdom and Germany.
The selected candidate will be hosted within the « Information, Multimodality & Signal » research group on the Metz campus of Supélec. The group is composed of approximately 30 persons involved in machine learning, signal processing and high-performance computing. During the ILHAIRE project, the IMS group is in charge of developing an interaction management system incorporating laughter. The IMS group also conducts researches in the domains of smart environments and robotics and the candidate will benefit from new facilities recently acquired by the campus for developing this research.
Profile
The candidate should hold a Master or engineering degree in electrical engineering, mathematics or computer science; have a good knowledge of machine learning and statistics as well as in C++ programming. Working knowledge of English is mandatory.
Contacts:
Olivier Pietquin
olivier.pietquin@supelec.fr
Matthieu Geist
matthieu.geist@supelec.fr
IMS homepage: ims.metz.supelec.fr
PhD Position [3y; MSc or engineering degree in electrical engineering, mathematics or computer science; knowledge of machine learning & statistics; C++; English; Imitation Learning and Task Transfer for Natural Human-Computer Interaction] / Supélec - IMS Research Group; Metz
علاقه مندی ها (Bookmarks)