This seminar aims to increase the links between the different laboratories in Saclay in the field of Applied Maths, Statistics and Machine Learning. The Seminar is organized every first Tuesday of the month with 2 presentations followed by a small refreshment. The localization of the seminar will change to accommodate the different labs.
Organization
Due to access restriction, you need to register for the seminar. A link is provided in the description and should also be sent with the seminar announcement. It will also help us organize for the food quantities. If you think you will come, please register! (even if you are unsure)
To not miss the next seminar, please subscribe to the announcement mailing list palaisien@inria.fr.
You can also add the calendar from the seminar to your own calendar (see below).
Next seminars
REGISTER
07 Apr 2026, 12h At Inria Saclay - Amphi Sophie Germain
Christophe Kervazo-Combining deep learning and optimization for remote sensing inverse problems
Remote sensing images are becoming increasingly useful in many applicative fields such as geology, environment monitoring, and urban planning. Among existing modalities, hyperspectral images benefit from a very high spectral resolution, enabling in principle an easy identification of the materials which are present within the scene. Nevertheless, due to physical limitations, such a high spectral resolution comes at the expense of a low spatial resolution on ...
Remote sensing images are becoming increasingly useful in many applicative fields such as geology, environment monitoring, and urban planning. Among existing modalities, hyperspectral images benefit from a very high spectral resolution, enabling in principle an easy identification of the materials which are present within the scene. Nevertheless, due to physical limitations, such a high spectral resolution comes at the expense of a low spatial resolution on the ground (e.g. 30m x 30m per pixel), complicating image interpretation.
During this presentation, we will explore two complementary ways of bypassing this issue, namely hyperspectral unmixing and hyperspectral super-resolution through data fusion. The algorithms we will discuss are both based on algorithm unrolling, a paradigm enabling to learn, from a training set, some acceleration factors within an iterative optimization algorithm. The advantage of such an approach is twofold: on the one hand, compared to traditional iterative methods, the results are more accurate while the number of iterations dramatically decreases. On the other hand, the interpretability is improved compared to black-box neural networks. In addition, we will also present experimental evidence demonstrating that our unrolled algorithms, trained on simple synthetic datasets, generalize effectively to real-world datasets during testing.
Particle Swarm Optimization (PSO) is a class of heuristic methods for gradient-free global minimization. Despite their popularity, the reasons for their sucess remain poorly undernstood. In this talk, I will expose some of the theoretical work in that direction, starting from the mean-field approximation of Pinnau et al. (2016), to our recent work on long-time convergence. In particular, we show that a simplified version of CBO follows, in expectation, a dynamic similar to that of ...
Particle Swarm Optimization (PSO) is a class of heuristic methods for gradient-free global minimization. Despite their popularity, the reasons for their sucess remain poorly undernstood. In this talk, I will expose some of the theoretical work in that direction, starting from the mean-field approximation of Pinnau et al. (2016), to our recent work on long-time convergence. In particular, we show that a simplified version of CBO follows, in expectation, a dynamic similar to that of proximal gradient descent. This allows to highlight two phases in the PSO dynamics : a global search phase based on a proximal map followed with a local Gaussian random search.
Joint work with Victor Priser and Pascal Bianchi
REGISTER
12 May 2026, 12h At Inria Saclay - Amphi Sophie Germain
Intelligent robots do not just respond to commands; they imagine what you meant, what you wanted, what you believed. And they do this while learning from very little, and running on a chip in your living room. In this talk, I will present recent advances in generative modeling that aim to equip embodied agents with efficient models that can run faster, with fewer data, and more efficient models and imagine possible futures under uncertainty.
Intelligent robots do not just respond to commands; they imagine what you meant, what you wanted, what you believed. And they do this while learning from very little, and running on a chip in your living room.
In this talk, I will present recent advances in generative modeling that aim to equip embodied agents with efficient models that can run faster, with fewer data, and more efficient models and imagine possible futures under uncertainty.
Scientific Committee
The program and the organization of this seminar is driven by a scientific committee composed of members of the different laboratories in Saclay. The members of the committee are currently:
