Research project
I had the chance to pursue this year at Télécom Paris in the IDS department (Information, Data and Signal) a research program, which was a selective.
During this program, I work with Prof. Pascal Bianchi on developing the theoretical part of the Stein Variational Gradient Descent (SVGD) which is a recently discovered algorithm (since 2016) that is used to sample from a wide variety of probability distributions because it does not require some very restrictive conditions on the distribution to sample from.
This algorithm can be used in a wide variety of Machine Learning tasks such as Reinforcement learning and Bayesian Machine Learning. Moreover, the task of sampling from a distribution is so important to do simulations in a lot of domains like Biology if we want to simulate the propagation of a disease in a group, or in Finance as well to use Monte Carlo methods (approximating integrals with discrete sums).
This algorithm uses a “kernelized” version of the Wasserstein Gradient Descent algorithm, more information about this algorithm can be found in my state of the art.
You will find here my final article, where I propose new algorithms that can improve SVGD. The code for simulations can be found in my Github repository for SVGD.
Garden connection kit: PACT'AG
During our first year at Télécom Paris, we created as a team of 6 members, a kit to connect a garden.
If you want to start growing your own fruits and vegetables, or any plant that you want, then you can use our product (this is just the purpose of the project, but we are not selling it actually !).
The kit is comprised of a box that contains micro-controller that controls the actuators like lamps and pumps, and get the values of the different sensors like temperature, humidity and luminosity.
To manipulate the garden, like activating the pumps or just getting the temperature of the garden, we developed an android application that communicates with the micro-controller using a server. This server modifies a database and send the client’s orders to the micro-controller to execute them. Here is the git repository of our work: PACT’AG.
You can also check this nice poster that a friend has made for the project: poster
Stereogame 🥎🏏
As a team of 4, we created a game using OpenGL and the Python library pygame, which we called stereogame.
It is a two players game that looks like tennis but where you use a bat instead of a racket. And there are some special effects like explosive balls, or cloning the ball into 3, or teleportation, etc.
And what makes this game special is that it is suited for a 3D TV, and both players look at the same TV but from different angles, and when you change the angle you can see the camera of the other player. So using this technique, we can add as many players as we want in the game, and they will all see their own camera in full screen using on single TV. Hence, we don’t need to split our screen to sub-screens !
See the git repository of this project.
