Spoken Dialogue Systems

My main area of research is machine learning for optimising man-machine interaction strategies and especially reinforcement learning for optimising spoken dialogue management strategies. I mostly worked on user simulation as well as batch and online learning of dialogue strategies with the originality of mixing research in theoretical machine learning and human-machine interaction. After some years spent on this topic we came up with algorithms able to learn online an optimal policy while interacting with users. I collaborated with the main actors in this domain and was involved in one of the first EU project dedicated to this problem: the FP7 ICT CLASSiC project. You can find a non exhaustive list of my publications about spoken dialogue systems here.

Reinforcement Learning

My research in machine learning for interaction strategies optimisation led me to go deeper into reinforcement learning theory. Especially, together with Matthieu Geist, we investigated value function approximation schemes based on Kalman filtering and statistical linearization. These schemes conducted to online and sample efficient algorithms able to deal with non stationarity and allowing for uncertainty management which are highly desirable features in human-machine interaction. You can find a selected list of publications here. See my full publication list for other papers.

Inverse Reinforcement Learning

Reinforcement learning is a paradigm where a machine learns to behave because it optimizes a cumulative function of rewards it gets after each action it takes. The advantage of this approach is that it requires little prior knowledge about the environment and thus avoids modeling biases. Yet, defining the reward to optimize is a challenge in it own right and requires prior knowledge. This is why I investigated the field of inverse reinforcement learning (IRL), where a learning agent infers the reward optimizing by an expert demonstrating the task. Again, this is of great interest in human-machine interaction since, considering humans as experts in interaction, data of human-human interaction can serve to learn rewards. See a selected list of publications on IRL here among which NIPS, IJCAI and ECML papers.

Other Interactive Systems

In addition to spoken dialogue systems, I'm also interested in optimizing human-machine interaction in general thanks to (direct or inverse) reinforcement learning. This is why I also worked in the fields of automotion within the FP7 SST ISi-PADAS EU project, tutoring in the ALLEGRO INTERREG IVa project and I'm now involved in the FET OPEN ILHAIRE project about laughing machines and I'm the coordinator of the ANR ContInt MaRDi project about human-robot interaction. Some publications can be found here.

Biomedical Signal Processing

Besides my main activity in human-machine interaction, I've been much interested in biomedical signal processing by means of machine learning and especially in ECG signals and fMRI imaging. Most of this work has been done thanks to a fruitful collaboration with the IADI lab and continues with the Oxford Institute of Biomedical Engineering. See publications here.

Speech Processing

Because I did my PhD thesis in a speech lab (TCTS lab of the Faculté Polytechnique de Mons), I naturally got interested in speech processing. Some publications are here.

PhD Students

  1. Matthieu Geist
  2. Julien Oster
  3. Béatrice Chevaillier
  4. Sarah Ménétré
  5. Senthilkumar Chandramohan
  6. Lucie Daubigney
  7. Edouard Klein
  8. Bilal Piot
  9. Layla El Asri
  10. Timothé Collet
  11. Hadrien Glaude
  12. Alexandre Bérard
  13. Julien Perolat
  14. Merwan Barlier
  15. Nicolas Carrara
  16. Florian Strub

Ongoing Funded Projects

Acronym Project Name Funding agency Dates
BabyRobot Child-Robot Communication and Collaboration: Edutainment, Behavioural Modelling and Cognitive Development in Typically Developing and Autistic Spectrum Children 2016-2019
IGLU Interactive Grounded Language Understanding 2015-2018
KEHATH Advanced Quality Methods for Post-Edition of Machine Translation 2014-2018
MaRDi Man-Robot Dialogue 2012-2016

Past Funded Projects

Acronym Project Name Funding agency Dates
Incorporating Laughter into Human Avatar Interactions: Research and Experiments. 2010-2013
Méthodologie d’évaluation des algorithmes d’exploitation des enregistrements de la vidéoprotection 2011-2013
Adaptive Language LEarning technology for the Greater Region 2010-2013
Computational Learning in Adaptive Systems for Spoken Conversation 2008-2011
Integrated Human Modelling and Simulation to Support Human Error Risk Analysis of Partially Autonomous Driver Assistance Systems 2008-2011
Medi@tic 2008-2010
Large scale distributed INDexation of multimedia Objects 2007-2010
The European research taskforce creating human-machine interfaces SIMILAR to human-human communication 2003-2007

Template by OS Templates