For a brief introduction of myself, please click here. I also have a blog, where I write about whatever topics interest me enough to write about. This is where I explore ideas that are outside the scope of my research.
My employer is Inria Rennes in France and my research is focused on “Learning Algorithms for Inverse Problems with New Imaging Modalities“, which is the title of my doctoral thesis.
This means I try to reconstruct good, clean images from degraded observations, using algorithms that use machine learning.
Current focus
Currently, I am looking at deep equilibrium models (DEQ) for image reconstruction. DEQ use math to facilitate learning for architectures that are based on repeated application of neural nets. In essence, this means that DEQ can help us train iterative procedures end-to-end. I am currently attempting to leverage the advantages of DEQ to improve current algorithms for image reconstruction.
One future topic I will look at is extending current theory on image reconstruction to novel imaging modalities like omni-directional images or light-fields. The advantages of using DEQ (i.e. lower memory-usage when compared to unrolled methods) may become even more relevant when applied to large data samples and higher dimensional data.
Secondment
My secondment at Mid Sweden University has started this January (i.e. January 2023). Here, I am currently working on further improving existing methods for plug-and-play image reconstruction as well as photo-realistic image-generation and sampling from a prior distribution. I plan to begin extending my work to light-field reconstruction while I am here.