The ELISIR program recruits exceptional young talents directly after their Ph.D. to run a small team of scientists pursuing innovative independent research in the Life Sciences.
\r\nFeel free to attend this presentation (in SV 1717 or virtually via the zoom link) !
\r\n
Abstract:
\r\nOlfactory navigation emerges from dynamic interactions among sensory input, brain activity, and the animal’s own movements. In C. elegans, a compact nervous system enables brain-wide imaging in behaving animals exposed to controlled olfactory environments, and undulatory locomotion reduces navigation to transitions among forward runs, reversals, and turns. Here, we used brain-wide tracking microscopy to record neural activity in freely crawling worms navigating spatial odor gradients or isotropic environments, and in immobilized worms responding to temporal odor pulses. Across conditions, we identified strongly correlated brain-wide activity patterns spanning olfactory sensory neurons, interneurons, and premotor neurons. Motor-state transitions and brain-wide correlation structure differed between freely moving and immobilized animals, demonstrating that olfactory sensory representations depend jointly on sensory input, motor activity, and behavioral context. Together, these findings show that brain-wide representations of olfactory behavior are extensively reconfigured by sensory input, self-motion, and behavioral context.
\r\n
The 2026 Journée Georges de Rham will take place on Wednesday 03 June 2026 in the Polydôme at EPFL, from 15:00 to 17:30 followed by an apéritif.
\r\n
\r\nProfessor Wang will deliver a talk titled “Kakeya sets in R^3″. A Kakeya set is a compact subset of R^n that contains a unit line segment pointing in every direction. Kakeya set conjecture asserts that every Kakeya set has Minkowski and Hausdorff dimension n. We prove this conjecture in R^3 as a consequence of a more general statement about union of tubes. Joint work with Josh Zahl.
\r\n
\r\nProfessor Candès will deliver a talk titled “What Statistics and AI Offer Each Other?“, exploring how thinking carefully about AI inputs and outputs yields more powerful, safer AI. By examining several vignettes, we shall answer questions such as: how do we train language models under cost constraints? What happens when we’ve exhausted all available data? If I start a clinical trial using the drug AI thinks is best, will it pan out? How can we ensure high-quality products when AI is used in a larger workflow? That is, how do I know whether AI automated a task correctly? AI powered imputations are beginning to substitute for real data when collection of the latter is difficult, slow, or costly. How then should we leverage machine learning predictions both as a substitute for high-quality data and as a tool for guiding real data collection?
\r\n
\r\nBernoulli Center for Fundamental Studies
\r\n
First-order methods are widely used in optimization and machine learning, and their behavior is often analyzed through the spectrum of worst case convergence rates. Obtaining such guarantees is often difficult and both time consuming and error-prone. Starting with the work of Drori and Teboulle (2014), novel techniques have been used to gain numerical insights, leading to the release of various performance estimation (PE) software.
\r\n
\r\nIn this talk, I will show how various computer-aided techniques can be used to study first-order optimization methods in a systematic way. From performance estimation problems with automated Lyapunov discovery, to symbolic regression and computer algebra systems, novel tools completely reshape the way we approach theory of optimization.
\r\nAs a main example, I will focus on error feedback methods used with compressed communication in distributed optimization. While error feedback has been widely studied, existing theory often provides untight (thus unreliable) bounds. I will present tight analyses with matching lower bounds that allow a fair comparison between error feedback schemes and standard compressed gradient descent, and help explain when error feedback is useful and when it is not.
\r\n
\r\nOverall, the talk aims to show how various computer-aided proofs can lead to clearer and more reliable insights into first-order optimization methods.
\r\n
\r\nBased on :
\r\nA Tight Theory of Error Feedback Algorithms in Distributed Optimization, DB Thomsen, A Taylor, A Dieuleveut
\r\nInternational Conference on Machine Learning (ICML 2026)
\r\nTight analyses of first-order methods with error feedback, DB Thomsen, A Taylor, A Dieuleveut
\r\nAdvances in Neural Information Processing Systems (NeurIPS) (2025).
\r\nPEPit: computer-assisted worst-case analyses of first-order optimization methods in Python
\r\nB Goujaud, C Moucer, F Glineur, JM Hendrickx, AB Taylor, A Dieuleveut
\r\nMathematical Programming Computation 16 (3), 337-367 (2024)
\r\n(eventually - as a detour) Provable non-accelerations of the heavy-ball method B Goujaud, A Taylor, A Dieuleveut, Mathematical Programming, 1-59 (2025)
\r\n
The ELISIR program recruits exceptional young talents directly after their Ph.D. to run a small team of scientists pursuing innovative independent research in the Life Sciences.
\r\nFeel free to attend this presentation (in SV 1717 or virtually via the zoom link) !
Thesis Director: Prof. J.-Ph. Brantut,
\r\nPhysics doctoral program
\r\nThesis Nr. 11525
\r\n
\r\nTo take part in the public defense, please contact directly the speaker