Institut de recherche mathématique avancée

Résumé : Abstract: We study operator learning in the context of linear propagator models for optimal order execution problems with transient price impact a la Bouchaud et al. (2004) and Gatheral (2010). Transient price impact persists and decays over time according to some propagator kernel. Specifically, we propose to use In-Context Operator Networks (ICON), a novel transformer-based neural network architecture introduced by Yang et al. (2023), which facilitates data-driven learning of operators by merging offline pre-training with an online few-shot prompting inference. First, we train ICON to learn the operator from various propagator models that maps the trading rate to the induced transient price impact. The inference step is then based on in-context prediction, where ICON is presented only with a few examples. We illustrate that ICON is capable of accurately inferring the underlying price impact model from the data prompts, even with propagator kernels not seen in the training data. In a second step, we employ the pre-trained ICON model provided with context as a surrogate operator in solving an optimal order execution problem via a neural network control policy, and demonstrate that the exact optimal execution strategies from Abi Jaber and Neuman (2022) for the models generating the context are correctly retrieved. Our introduced methodology is very general, offering a new approach to solving optimal stochastic control problems with unknown state dynamics, inferred data-efficiently from a limited number of examples by leveraging the few-shot and transfer learning capabilities of transformer networks.

Résumé : The numerical approximation of partial differential equations (PDEs) plays a crucial role in various fields, including engineering, mechanics and physics, for design and assessment. To accurately account for uncertainty in parameter values, we must solve the numerical model for a wide range of relevant parameters; an efficient numerical solution of this type of problem is even more challenging in a real time context. Model order reduction (MOR) methods have the purpose to overcome the computational obstacle of numerical simulations to large-scale dynamical systems. I will show linear and nonlinear solution manifold approximations and their validity for different applications in radioactive waste management and in contact mechanics. Keywords: reduced basis method, Galerkin approximation, domain decomposition, supervised machine learning.

Résumé : Les méthodes Hybrid High-Order (HHO) [Di Pietro et al.], constituent un cadre numérique à la fois flexible et robuste pour la résolution de problèmes aux dérivées partielles. Ce séminaire propose une introduction accessible à ces méthodes, en mettant l’accent sur leurs fondements théoriques — consistance, stabilité et convergence — illustrés par le problème de Poisson (-Δu = f) sur un domaine simple comme le carré. Nous étendrons ensuite l’analyse au cas du transport-réaction, en considérant par exemple le problème d’Oseen. Les connexions avec d’autres méthodes numériques seront également abordées, en comparant HHO aux éléments finis classiques et aux méthodes discontinues de Galerkin (dG), tout en esquissant, si le temps le permet, des liens avec des approches plus récentes telles que les méthodes hybrides de Galerkin discontinu (HDG) et les méthodes d’éléments virtuels (VEM). Enfin, nous discuterons des aspects pratiques d’implémentation, en particulier l’exploitation du calcul parallèle, pour mettre en lumière les atouts de HHO dans des contextes computationnels plus exigeants.

Résumé : In this talk we will explain a computation describing Hochschild-Kostant-Rosenberg isomorphism theorems as exponential maps. This computation uses the construction of a filtered circle obtained in collaboration with Moulinos and Toën. As two independent applications we will discuss a definition of motivic Donaldson-Thomas invariants in positive characteristic and the extension of Hochschild homology for elliptic curves.