Thibaut Kulak

Short Biography

I am a PhD student at Idiap since December 2017, in the Robot Learning and Interaction Group.

I am part of the ROSALIS project (Robot skills acquisition through active learning and social interaction strategies). Motivated by the fact that most robot learning frameworks are turned towards learning from the available training data, in the ROSALIS project the goal is to take into account the social dimension of the learning process, and to learn more efficiently robot skills by having the robot using active learning to exploit as much as possible the interaction with the human. We think such a learning process will be more realistic because interactive, and will make possible to learn more complex skills as the data will be gathered intelligently through active learning strategies.


SoftBank Robotics Europe, Research Intern (April-Oct 2017)

When interacting with its environment, a robot must be able to predict the consequences of his actions. The goal of my internship in the AI Lab of Softbank Robotics was to use Recurrent Neural Networks to predict the future values of the robots's sensors given the robot's actual sensor values and its motor commands. I worked on how to learn latent variables from the environment to improve the prediction. Learning those latent variables is unsupervised, but variables like position of the robot can emerge naturally, similarly to the rat's place cells for instance.

I wrote a research article at the end of this internship, which has been accepted at ICANN 2018, but is for now only available on ArXiv :

Safran Morpho, Research Intern (Oct 2016 - March 2017)

Included in my formation at Ecole Centrale Paris, I worked one day every week on a research project at Safran Morpho during my last year of studies. The research project was on Head Pose Estimation with Deep Learning methods. The bottleneck of this approach is that it requires lots of data, which is expensive to obtain, that's why I worked on using unlabeled images to improve the model.

The report of the project is available on :


MVA master (M2), Ecole Normale Superieure (2016-2017)

Coursework :

- Deep Learning
- Object Recognition and Computer Vision
- Reinforcement Learning
- Graphs in Machine Learning
- Sparsity and Compressed Sensing
- Kernel Methods for Machine Learning
- Advanced Learning for Text and Graph Data
- Neurosciences
- Statistical Computing on Manifolds
- Large Scale Distributed Optimization

Some of the research projects I worked on for those classes are available on :

CentraleSupelec, Master's degree (2014-2017)

Data Science Specialization

Coursework :

- Machine Learning
- Statistics
- Data Mining
- Deep Learning
- Natural Language Processing
- Sparsity and Compressed Sensing
- Statistical Computing on Manifolds
- Software Engineering

The University of Texas at Austin (Spring 2016)

I made a study-abroad to the University of Texas at Austin during Spring 2016, where I was in the Statistics and Data Sciences department.

Coursework :

- Monte-Carlo Methods in Statistics
- Design and Analysis of Experiments
- Mathematical Methods for Statistical Analysis
- Computational Bio and Bio-Informatics

Tel: +41 27 721 7815
Office: 201-5