À : Salle E244 Conf 4 24 rue Lhomond, 75005 Paris , à 14:00

À : Amphithéâtre Darboux 11 Rue Pierre et Marie Curie, 75005 Paris , à 14:00

À : , à 14:00

À : IV Conf IV, 24 rue Lhomond, 75005, Paris , à 14:00

À : , à 14:00

À : amphithéâtre Claude Bloch, Orme des Merisiers, bâtiment 772, CEA Saclay , à 14:00

We perform spectroscopy of resistively shunted Josephson junctions in a superconducting circuit QED experiment. We investigate a range of parameters for the junction impedance EJ/EC ≈ 0.1-10, covering the Cooper-pair box to the light transmon. The galvanically connected resistive shunt R is varied over several orders of magnitude, R ≈ 0.01-100 RQ, where RQ = h/(2e)2 ≈ 6.5 kΩ. Our setup avoids issues of previous experiments including Joule heating, DC noise, and implementation of a resistor as a finite transmission line. The data hints at a new phase diagram for the damped Josephson junction beyond the prediction by Schmid-Bulgadaev of an insulating state for R≫RQ. Our phase diagram approaches that of the damped harmonic oscillator for EJ≫EC and deviates from it for EJ≤EC. We clarify the role of a dissipative high impedance element on the dynamics of a Josephson junction in the quantum regime and set the stage for a better understanding of phenomena such as dual Shapiro steps, also known as Bloch oscillations.

À : : Salle de conference 3 rue d’ulm , à 14:00

À : Salle Conférence 4 (provisoire) 24 Rue Lhomond, Site Ulm, 75005, Paris , à 14:30

À : Collège de France , à 14:00

First, I introduced ZAugNet, a self-supervised deep learning method for z-slice augmentation in 3D microscopy. Biological 3D imaging often suffers from low z-resolution due to limitations in imaging techniques or sample properties. ZAugNet improves resolution by performing nonlinear interpolation between slices, effectively doubling z-resolution with each iteration. It outperforms existing methods across modalities and samples, using a GAN architecture and knowledge distillation for high-speed, accurate inference. I also developed ZAugNet+, an extension supporting continuous interpolation at arbitrary slice spacings. Both tools are open-source and available via a user-friendly PyTorch/Colab interface. In the second phase, I developed VertAX, a JAX-based computational framework for vertex model simulation and bilevel optimisation in confluent tissues. These tissues exhibit complex macroscopic behaviors governed by microscopic cell parameters. VertAX allows users to define custom energy and cost functions on vertex-based meshes and compute gradients with respect to any parameter using automatic differentiation. This enables efficient inverse modelling to match target behaviors or fit real data. We also benchmark equilibrium propagation as an alternative gradient estimation method for non-differentiable frameworks.

À : amphithéâtre Jacques-Louis Lions 48 Rue Barrault, bâtiment A, 75013, Paris , à 14:00

À : Salle L108 Mines Paris, 60 Bd Saint-Michel, 75272 Paris , à 14:00

À : : Amphithéâtre Durand 4 rue Jussieu, 75005 Paris Bâtiment Esclangon Sorbonne université, campus Pierre et Marie Curie , à 14:00