Elenco seminari del ciclo di seminari
“SEMINARI DI PROBABILITÀ”

I will introduce a continuous time probabilistic model for systems of interacting and spiking neurons. In this process, neurons spike at a rate depending on their membrane potential value. When spiking, they have a direct influence on their post-synaptic partners, namely, a fixed value, called "synaptic weight", is added to the potential of the postsynaptic neurons. In between successive spikes, due to some leakage effects, the membrane potential process follows a deterministic flow. Firstly, I will discuss the construction, well-posedness and the longtime behavior of the process, for a finite number of neurons and for infinite systems of neurons, both in the case with and without reset of the spiking neuron. I will then discuss mean field limits for the Hawkes description (without reset) of the model. In particular we will see how in the limit an ODE describing the evolution of the mean firing rate appears and how this approach allows to describe for example oscillatory behavior. If time permits, I will also discuss the influence of delay in the synaptic transmission and quickly speak about short term memory. The final part of the course will be devoted to the more difficult case with reset (the membrane potential of the spiking neuron goes back to a resting value, inducing discontinuities in the model). We will see how the limit process and its longtime behavior help us to explain important phenomena in neuroscience such as "metastability".

We will explore recent advances concerning nonlinear diffusion processes in the sense of McKean-Vlasov, and their connections to partial differential equations (PDEs) defined on the Wasserstein space, that is, the space of probability measures with finite second order moment. We will discuss recent results on the well-posedness - both in the weak and strong sense - of McKean-Vlasov stochastic differential equations (SDEs) driven by Brownian motion and/or jump processes. These results extend beyond the classical Cauchy-Lipschitz framework. In the Brownian setting, we will describe the regularization effect of the noise, notably the existence and smoothness of the transition density - particularly in the measure argument - under uniform ellipticity assumptions. These smoothing effects are crucial for establishing the existence and uniqueness of solutions to the Kolmogorov-type PDEs posed on the Wasserstein space, even in the presence of irregular terminal conditions and source terms. Such infinite-dimensional PDEs play a central role in deriving quantitative propagation of chaos estimates for mean-field approximations via interacting particle systems. Finally, we will discuss the numerical approximation of these equations using the Euler-Maruyama time discretization scheme.

We consider the generic divergence form second order parabolic equation with coefficients that are regular in the spatial variables and just measurable in time. We show that the spatial derivatives of its fundamental solution admit upper bounds that agree with the Aronson type estimate and only depend on the ellipticity constants of the equation and the L∞ norm of the spatial derivatives of its coefficients. We also study the corresponding stochastic partial differential equations and prove that under natural assumptions on the noise the equation admits a mild solution, given by anticipating stochastic integration.

We propose a novel small time approximation for the solution to the Zakai equation from nonlinear filtering theory. We prove that the unnormalized filtering density is well described over short time intervals by the solution of a deterministic partial differential equation of Kolmogorov type; the observation process appears in a pathwise manner through the degenerate component of the Kolmogorov's type operator. The rate of convergence of the approximation is of order one in the lenght of the interval. Our approach combines ideas from Wong-Zakai-type results and Wiener chaos approximations for the solution to the Zakai equation. The proof of our main theorem relies on the well-known Feynman-Kac representation for the unnormalized filtering density and careful estimates which lead to completely explicit bounds (joint work with Ramiro Scorolli)

We will discuss how the so-called product space framework, introduced in the '70s to represent delayed systems, can be effectively implemented to study well posedness of Kolmogorov-type PDEs with path-dependent coefficients. We will present results on linear and semilinear PDEs and as well as some related tools of stochastic calculus in Banach spaces.

In this talk we give a brief introduction to the theory of rough paths, that was developed by T. Lyons and studied by many others in the past decades. Then we discuss applications to McKean-Vlasov equations with common noise. Rough path theory builds on the theory of Young integration and aims to make sense of differential equations driven by a continuous path (but cadlag generalizations are possible) that need not be differentiable or of bounded variation. The driving signal can be Hölder continuous for any strictly positive Hölder exponent, in the talk we will only focus on path or regularity between 1/3 and 1/2. Many classical results of stochastic analysis can be recovered using rough path theory. We consider a McKean-Vlasov diffusion with "common" noise perturbed by a deterministic rough paths. These types of law-dependent equations arise as a mean-field limit of systems of interacting particles subject to a common noise that acts on each particle. Finally we give an outlook into applications to filtering theory.

In the context of randomly fluctuating interfaces in (1+1)-dimensions two Universality Classes have generally been considered, the Kardar-Parisi-Zhang and the Edwards-Wilkinson. Models within these classes exhibit universal fluctuations under 1:2:3 and 1:2:4 scaling respectively. We introduce a (1+1)-dimensional temperature-dependent model such that the classical ballistic deposition model is recovered as its zero-temperature limit. Surprisingly enough, its infinite-temperature version, the 0-Ballistic Deposition (0-BD) model, does not belong to either the universality classes mentioned above. We show that 0-BD has a scaling limit, a new stochastic process that we call Brownian Castle (BC) which, like any other renormalisation fixed point, is scale-invariant, in this case under the 1:1:2 scaling. The aim of the present talk is to provide a "global" construction of BC, determine some of its path-wise and distributional properties and prove its universality by showing that 0-BD converges to it. This is joint work with Martin Hairer (Fermat prize in 2013 and Fields medal in 2014)

This talk develops some aspects of stochastic calculus via regularization for processes with values in a general Banach space B. A new concept of quadratic variation which depends on a particular subspace is introduced. An Itô formula and stability results for processes admitting this kind of quadratic variation are presented. Particular interest is devoted to the case when B is the space of real continuous functions defined on [-T,0], T>0 and the process is the window process X(•) associated with a continuous real process X which, at time t, it takes into account the past of the process. If X is a finite quadratic variation process (for instance Dirichlet, weak Dirichlet), it is possible to represent a large class of path-dependent random variable h as a real number plus a forward integral which are explicitly given. This representation result of h makes use of a functional solving an infinite dimensional partial differential equation. This decomposition generalizes, in some cases, the Clark-Ocone formula which is true when X is the standard Brownian motion W. Some stability results will be given explicitly. This is a joint work with Francesco Russo (ENSTA ParisTech Paris).

Il seminario si baserà anche sul contributo: Corazza, G.E. & Lubart, T.A. (2019). Science and Method: Henri Poincaré. In Glaveanu, V. P. (Ed.). (2019). The creativity reader. Oxford University Press.

I will present probabilistic proofs of some regularity properties for the value function of general optimal stopping problems and for the associated optimal boundaries. In particular this talk focusses on C^1 regularity of the value function and Lipschitz continuity of the optimal boundary. Most of our arguments rely on fundamental concepts from the theory of Markov processes and bridge the probabilistic and the analytical strands of the literature on free boundary problems. I will also illustrate situations in which our work improves or complements known facts from PDE theory.

We consider a prototype class of Lévy-driven SDEs with McKean-Vlasov (mean-field) interaction in the drift. The coefficient is assumed to be affine in the state-variable and only measurable in the law. We study the equivalent functional fixed-point equation for the unknown time-dependent coefficients of the associated Markovian SDE. By proving a contraction property for the functional map in a suitable normed space, we infer existence and uniqueness results for the MK-V SDE, and derive a discretized Picard iteration method that approximates the law of the solution. Numerical illustrations show the effectiveness of the method, which appears to be appropriate to handle multi-dimensional settings. We finally describe possible extensions and generalizations to more general settings. This talk is based on joint work with Ankush Agarwal.

We consider an optimal control problem for piecewise deterministic Markov processes (PDMP) on a bounded state space. Here a pair of controls acts continuously on the deterministic flow and on the transition measure describing the jump dynamics of the process. For this class of control problems, the value function can be characterized as the unique viscosity solution to the corresponding integro-differential Hamilton-Jacobi-Bellman equation with a non-local type boundary condition. We are able to provide a probabilistic representation for the value function in terms of a suitable backward stochastic differential equation, known as nonlinear Feynman-Kac formula. The jump mechanism from the boundary entails the presence of predictable jumps in the PDMP dynamics, so that the associated BSDE turns out to be driven by a random measure with predictable jumps. Existence and uniqueness results for such a class of equations are non-trivial and are related to recent works on well-posedness for BSDEs driven by non quasi-left-continuous random measures.

The aim of this study is to present algorithms for the backward simulation of standard processes that are commonly used in financial applications. We extend the works of Ribeiro and Webber and Avramidis and L’Ecuyer on gamma bridge and obtain the backward construction of a Gamma process. Moreover, we are able to write a novel acceptance-rejection algorithm to simulate Inverse Gaussian (IG) processes backward in time. Therefore, using the time-change approach, we can easily get the backward generation of the Compound Poisson with infinitely divisible jumps, the Variance–Gamma the Normal–Inverse–Gaussian processes and then the time-changed version of the OU process (SubOU) introduced by Li and Linetsky. We then compare the computational costs of the sequential and backward path generation of such processes and show the advantages of adopting the latter one, in particular in the context of pricing American options or energy facilities like gas storages.

We recall some classic results on the regularity of solutions of stochastic differential equations. Then we consider two specific diffusion processes satisfying hypoellipticity conditions of Hormander type. Using Malliavin Calculus techniques recently developed to deal with degenerate problems, we find estimates for the density of the law of the solution, which we use to prove exponential bounds for the probability that the diffusion remains in a small tube, around a deterministic path, up to a given time. We then present some work in progress on asymptotic sharp estimates for the density and its derivatives for a similar, higher dimensional system.

Clark-Ocone formulas are powerful results in stochastic analysis with a variety of applications. In the talk we provide the Clark-Ocone formula for square-integrable functionals of point processes with stochastic intensity. Then we present two applications of the formula: the Poincare' inequality and a concentration bound for those functionals. Our results generalize the corresponding ones on the Poisson space. The talk is based on joint works with Ian Flint and Nicolas Privault (NTU, Singapore)

In the first part of the talk, mainly of the heuristic type, some basic notions (such as regularity, properness, disintegrability) are recalled and some examples are discussed. The second part is more technical and is devoted to some results and their implications. In the classical (Kolmogorovian) framework, a few 0-1 laws for regular conditional distributions are stated. Special attention is paid to the tail and the symmetric sigma- fields. In the coherent (de Finettian) framework, with reference to a Bayesian inferential problem, the existence of posterior distributions that make sufficient a given statistics, or make optimal a given estimator, is discussed. Finally, some compatibility problems for conditional distributions are mentioned, and a few asymptotic results are stated

In the first part of the talk, mainly of the heuristic type, some basic notions (such as regularity, properness, disintegrability) are recalled and some examples are discussed. The second part is more technical and is devoted to some results and their implications. In the classical (Kolmogorovian) framework, a few 0-1 laws for regular conditional distributions are stated. Special attention is paid to the tail and the symmetric sigma- fields. In the coherent (de Finettian) framework, with reference to a Bayesian inferential problem, the existence of posterior distributions that make sufficient a given statistics, or make optimal a given estimator, is discussed. Finally, some compatibility problems for conditional distributions are mentioned, and a few asymptotic results are stated

Il XX è stato, fra altre cose, anche il secolo della rinascita della probabilità e della statistica nelle comunità scientifiche più avanzate dell'occidente. Questo seminario si propone di illustrare il contributo degli italiani a tale processo, a partire dalle questioni che, polarizzando l'interesse di alcuni nostri studiosi di forte ingegno, portarono alla formulazione di teorie originali ed al conseguimento di pregevoli risultati destinati a durare nel tempo. Di essi si darà cenno, breve ma sperabilmente sufficiente a chiarirne valore e ruolo in relazione allo sviluppo generale delle scienze e dei metodi, insieme a qualche considerazione sulle caratteristiche umane e professionali degli Autori.

We introduce a simple class of mean field games with absorbing boundary over a finite time horizon. In the corresponding $N$-player games, the evolution of players' states is described by a system of weakly interacting It{\^o} equations with absorption on first exit from a bounded open set. Once a player exits, her/his contribution is removed from the empirical measure of the system. Players thus interact through a renormalized empirical measure. In the definition of solution to the mean field game, the renormalization appears in form of a conditional law. We justify our definition of solution in the usual way, that is, by showing that a solution of the mean field game induces approximate Nash equilibria for the $N$-player games with approximation error tending to zero as $N$ tends to infinity. This convergence is established provided the diffusion coefficient is non-degenerate. The degenerate case is more delicate and gives rise to counter-examples. This talk is based on a joint work with Markus Fischer (Università degli Studi di Padova).