Séminaire Statistique
organisé par l'équipe Statistique
-
Modou Wade
A general framework for deep learning
20 mars 2026 - 14:00Salle de séminaires IRMA
This paper develops a general approach for deep learning for a setting that includes nonparametric regression and classification. We perform a framework from a data that fulfills a generalized Bernstein-type inequality, including, independent, ϕ-mixing, strongly mixing, C-mixing observations. Two estimators are proposed: a non-penalized deep neural network estimator (NPDNN) and a sparse-penalized deep neural network estimator (SPDNN). For each of these estimators, bounds of the expected excess risk on the class of Hölder smooth functions and composition Hölder functions are established. Applications to independent data, as well as to ϕ-mixing, strongly mixing, C-mixing processes are considered. For each of theses examples, the upper bounds of the expected excess risk of the proposed NPDNN and SPDNN predictors are derived. It is shown that, both the NPDNN and SPDNN estimators are minimax optimal (up to a logarithmic factor) in many classical settings. -
Christelle Agonkoui
A venir
26 mars 2026 - 11:00A confirmer
-
Orlane Rossini
From Impulse Control of PDMPs to Bayesian Adaptive POMDPs: A Reinforcement Learning Approach
27 mars 2026 - 11:00Salle de séminaires IRMA
Piecewise Deterministic Markov Processes (PDMPs) constitute a family of Markov processes characterized by deterministic motion interspersed with random jumps. When controlled through discrete-time interventions, this leads to an impulse control problem. In the fully observed setting with known dynamics we develop a numerical method to compute an optimal strategy. In real-world applications, however, full observability is rarely available. Under partial observation, the impulse control of a PDMP can be reformulated as a Partially Observed Markov Decision Process (POMDP), which we address using deep reinforcement learning techniques. A major limitation of existing approaches is the assumption that the underlying PDMP dynamics are known or can be accurately simulated. This assumption is unrealistic in applications such as patient monitoring, where data may be scarce and disease dynamics may vary across individuals. To address this issue, we introduce a Bayesian Adaptive POMDP (BAPOMDP) framework, in which the unknown PDMP parameters are modeled probabilistically and updated through Bayesian inference. The resulting continuous-state BAPOMDP is solved using deep reinforcement learning methods adapted to high-dimensional belief spaces. This work thus combines stochastic control theory, Bayesian modeling, and deep reinforcement learning to provide a unified framework for decision-making under partial observability and model uncertainty. The proposed methodology is thoroughly illustrated and validated on a medical application : the adaptive follow-up and monitoring of patients diagnosed with multiple myeloma. -
Hugo Lebeau
A venir
3 avril 2026 - 11:00Salle de séminaires IRMA
-
Antoine Heranval
Analyzing temporal dependence between extreme events using point processes
10 avril 2026 - 11:00Salle de séminaires IRMA
Extreme meteorological events often occur in complex temporal configurations, where the impacts of one hazard may depend on the prior occurrence of others. Characterising such temporal dependencies is essential for understanding compound climate risks, yet remains challenging due to the discrete, heterogeneous, and clustered nature of extreme events. In this study, we apply temporal point process methods to characterise dependencies among extreme meteorological events occurring within appropriately defined spatial regions across Europe, focusing exclusively on their temporal structure.
We introduce an event-based framework in which extreme events are represented as marked temporal point processes, with marks describing key characteristics such as intensity or duration. Global first- and second-order temporal statistics are used to quantify clustering, co-occurrence, and directional dependencies between different types of extremes. In particular, we rely on directional cross-$K$ functions to assess whether the occurrence of one type of extreme event systematically modifies the short-term probability of subsequent events of another type.
Two complementary applications illustrate different facets of compound event analysis. First, we demonstrate the relevance of the framework for preconditioned compound events through a temporal analysis of wildfire-related meteorological extremes. Second, we examine temporal dependence between extreme precipitation, extreme wind, and extreme atmospheric instability across all European NUTS-2 regions.
Building on these second-order statistics, we develop formal tests of temporal independence to assess the significance of observed directional interactions between different types of extreme events. Overall, this temporal point process framework provides a rigorous and interpretable approach to the analysis of compound and preconditioned climate extremes, with direct applications to climate risk assessment and early-warning systems.