I’m a third-year PhD student in image processing at the CREATIS lab. I’m particularly interested in optimization, inverse problems, and deep learning.
The topic of my thesis is multiplicative algorithms (aka Richardson-Lucy) and their extension to the regularized setting, more particularly to regularizations defined through neural networks, often called Plug-and-Play. The goal of developing such methods is to improve image reconstruction and restoration in applications such as emission tomography (PET and SPECT) astronomy and microscopy.
The three contributions of my thesis are:
- an algorithm mixing multiplicative updates with diffusion models
- an algorithm mixing multiplicative updates with Plug-and-Play regularization
- and most importantly, a new Bregman proximal gradient algorithm which can be specialized to recover the multiplicative updates and extend them to any regularization
I’m also known as Deepia on YouTube, where I do animated videos about topics in deep learning.
Finally, I’m a maintainer of the DeepInverse library for solving inverse problems with deep learning.
Before my thesis I graduated from the Data & Intelligence for Smart Systems Master’s degree from the Computer Science department of Claude Bernard Lyon 1 University. I’ve done a 6 months research internship at Thales for my Master’s thesis, focusing on deep learning applied to subresolved infrared target detection.