Kellian Cottart

Kellian Cottart

Doctoral student in Bayesian AI and Neuromorphic engineering at Université Paris-Saclay | Data Sciences at Ecole Polytechnique (IPP2023) | Engineer's degree at Ecole Centrale d'Electronique

Sitemap

A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.

Pages

Posts

A geometrical interpretation of variational methods in machine learning

6 minute read

Published:

Variational methods are often introduced through information theory: learning means finding an approximate distribution that remains close to a target Bayesian posterior. This has a geometrical interpretation. The mathematical object being optimized is not a point in a Euclidean parameter space, it's a probability distribution living on a curved statistical manifold.

Itinérances - Ph.D Student Portrait - Lara Couronné

2 minute read

Published:

This portrait celebrates Lara Couronné, a brilliant scientist whose journey blends determination, passion, and creativity. Filmed as part of her daily life in research, this video offers a glimpse into who she is, both in and beyond the lab.

Bayesian Continual Learning in Nature Communications

1 minute read

Published:

I’m super happy to announce that Djohan Bonnet, Damien Querlioz and I, as well as co-authors recently published a paper about Bayesian continual learning & forgetting in Neural Networks in Nature Communications! It’s been a long journey to get this paper published and we are very proud of our research.

Itinérances - Ph.D Student Portrait - Kellian Cottart

1 minute read

Published:

Itinérances is a portrait of my journey as a Ph.D. student, at the crossroads of scientific research, personal passion, and intellectual curiosity. This short film explores my daily life in the lab, the motivations that drive my work, and the ideas that shape my approach to science.

gallery

publications

A unified method to generate representative volume elements with tailored random fibre arrangements to estimate the shear and transverse behaviours of unidirectional continuous fibres composite plies

Published in Journal Of Composite Materials, 2024

Different microstructures with or without clusters of fibres (resulting from manufacturing processes) are necessary to quantify the mechanical variability. The variability of the microstructures is characterised by the variation of the arrangement, the volume fraction and the misalignment of the fibres.

Recommended citation: Mechin P-Y, Borras A, Cottart K, Keryvin V. A unified method to generate representative volume elements with tailored random fibre arrangements to estimate the shear and transverse behaviours of unidirectional continuous fibres composite plies. Journal of Composite Materials. 2024;0(0). doi: 10.1177/00219983241300144
Download Paper | Download Slides

Dynamic Control of Weight-Update Linearity in Magneto-Ionic Synapses

Published in Nanoletters, 2025

We demonstrate that magneto-ionic devices can perform as synaptic elements with dynamically tunable depression linearity controlled by an external magnetic field, a functionality reminiscent of neuromodulation in biological systems.

Recommended citation: Bernard, Guillaume, et al. "Dynamic control of weight-update linearity in magneto-ionic synapses." Nano Letters 25.4 (2025): 1443-1450. doi: 10.1021/acs.nanolett.4c05247
Download Paper | Download Slides

Bayesian continual learning and forgetting in neural networks

Published in Nature Communications, 2025

We introduce Metaplasticity from Synaptic Uncertainty (MESU), a Bayesian update rule that scales each parameter’s learning by its uncertainty, enabling a principled combination of learning and forgetting without explicit task boundaries.

Recommended citation: Bonnet, Djohan, et al. "Bayesian continual learning and forgetting in neural networks." Nature Communications 16.1 (2025): 9614.
Download Paper | Download Slides

talks

Neuromorphic Symposium: Bayesian Continual Learning and Forgetting in Neural Networks (15 minutes)

2 minute read

Published:

When neural networks are deployed in the real world, data does not stay stationary. Environments evolve, tasks change, and distributions drift. Standard neural networks suffer from catastrophic forgetting: when learning something new, they overwrite what they learned before. But there is a second, often ignored challenge: catastrophic remembering, when networks become too rigid, stop adapting, and lose their ability to fit new data, leading to overconfident predictions.

teaching