Major: Computer Science, Big Data and connected objects
ESILV is a generalist and digital engineering school. Located in the Pole Universitaire Leonard de Vinci.
Middle School Diploma, with first class honours, 7th in the national ranking. (Official Note)
Scientific Baccalaureat, with first class honours
In this project, my comrades Thibaut Tucoulou, Benoit Feurgard, Paul Thibonnier and myself wanted to develop a computer vision prototype, capable of detecting everyday objects in a room, guiding an individual vocally towards the objects he wants to reach.
1st prize in the Défi H competition, we succeeded in convincing people of the importance and usefulness of such a system.
Proposal of a deep network architecture for the generation of sports commentary stream summaries. Publication: EGC 2020 TextMine.
I have always enjoyed discussing Machine Learning, Deep Learning, Mathematics with some friends, explaining to one another the brand new architectures, the improvements, trying to understand in depth what we learn etc. I then decided to launch my own blog on which I will be talking about papers that I really enjoyed reading, but also breaking through ML theory in an intuitive and enjoying way.
In recent years, we have seen an influx of a large number of high-performance neural models in NLP, but with astronomical dimensions. In this project that I started in April 2020, I study the latest state of the art models in the field of automatic summarization; I then try to build smaller models with the objective of minimizing performance loss. The first results are quite conclusive.
"ENTROPY-SGD: BIASING GRADIENT DESCENT INTO WIDE VALLEYS" is a paper published in 2017 in which the authors try to explain generalization in deep learning under certain assumptions. It is a paper that I liked very much, so I decided to write a blog article on it and to implement an optimization method according to the algorithm they propose.
Re-implementation of a neural network architecture for automatic summarization.
My mission in this project is to implement a neural architecture combining the "Pointer-Generator" and "Transformer" models for the automatic generation of abstract summaries.
During my courses on 3D Vision, I had the opportunity to address the subject of 3D rendering, as well as a study on the equation of light. I was thus able to implement mathematical models to simulate the exposure to light of certain objects, their brightness, color, texture etc.
