We are excited to announce the upcoming Stochastic Numerics and Statistical Learning: Theory and Applications Workshop 2024, taking place at KAUST, Auditorium 0215 b/w B4&5, from May 19-30, 2024. Following the highly successful last two years edition, this year's workshop promises to be another engaging and insightful event for researchers, faculty members, and students interested in stochastic algorithms, statistical learning, optimization, and approximation. The 2024 workshop aims to build on the achievements of last year's event, which featured 30 talks, two mini-courses, and two poster sessions, attracting over 150 participants from various universities and research institutes. In 2022 and 2023, attendees had the opportunity to learn from through insightful talks, interactive mini-courses, and vibrant poster sessions. This year, the workshop will once again showcase contributions that offer mathematical foundations for algorithmic analysis or highlight relevant applications. Confirmed speakers include renowned experts from institutions such as Ecole Polytechnique, EPFL, Université Pierre et Marie Curie - Paris VI, CUHK Shenzhen, and Imperial College London, among others.

Dear Kaustians, We are excited to announce the upcoming Stochastic Numerics and Statistical Learning: Theory and Applications Workshop 2023, taking place at KAUST, Building 9, from May 21 to June 1, 2023. Following the highly successful 2022 edition, this year's workshop promises to be another engaging and insightful event for researchers, faculty members, and students interested in stochastic algorithms, statistical learning, optimization, and approximation. The 2023 workshop aims to build on the achievements of last year's event, which featured 28 talks, two mini-courses, and two poster sessions, attracting over 150 participants from various universities and research institutes. In 2022, attendees had the opportunity to learn from through insightful talks, interactive mini-courses, and vibrant poster sessions. This year, the workshop will once again showcase contributions that offer mathematical foundations for algorithmic analysis or highlight relevant applications. Confirmed speakers include renowned experts from institutions such as Ecole Polytechnique, EPFL, Université Pierre et Marie Curie - Paris VI, and Imperial College London, among others.

This talk focuses on a 'particle MCMC' method known as the conditional particle filter (CPF), or the particle Gibbs. The CPF is a slight algorithmic variant of the original particle filter, but serves a different purpose: it defines an MCMC transition targeting the HMM smoothing distribution. The empirical evidence suggests that certain variants of the CPF mix well even in high dimensions (with long observation records). We review some theoretical insights that consolidate such empirical findings, and justify why the CPF is often efficient for HMM inference.

This thesis focuses on the use of multilevel Monte Carlo methods to achieve optimal error versus cost performance for statistical computations in hidden Markov models as well as for unbiased estimation under four cases: nonlinear filtering, unbiased filtering, unbiased estimation of hessian, continuous linear Gaussian filtering.

It is now fairly common to use Sequential Monte Carlo (SMC) algorithms for normalizing constant estimation of high-dimensional, complex distributions without any particular structure. In order for the algorithm to give reasonable accuracy, it is well known empirically that one must introduce appropriate intermediate distributions that allow the particle system to “gradually evolve” from a simple initial distribution to the complex target distribution, and one must also specify an appropriate number of particles to control the error. Since both the number of intermediate distributions and the number of particles affect the computational cost of the algorithm, it is crucial to study and attempt to minimize the computational cost of the algorithm subject to constraints on the error.

Filtering (or data assimilation) is the process of combining noisy observational data with an output of a numerical model to produce an optimal estimate of the evolving state of the system.

Many alternatives to the probabilistic modelling of information have been proposed since the birth of modern Statistics; yet, few have been successfully applied to the complex inference problems that modern Statisticians are faced with.

Bayesian inference for nonlinear diffusions, observed at discrete times,is a challenging task that has prompted the development of a number of algorithms, mainly within the computational statistics community. We propose a new direction, and accompanying methodology - borrowing ideas from statistical physics and computational chemistry - for inferring the posterior distribution of latent diffusion paths and model parameters, given observations of the process

Fluid dynamics models are ubiquitous in a multitude of applications. One of the most important applications of fluid dynamics models is numerical weather prediction. Modern numerical weather prediction combines sophisticated nonlinear fluid dynamics models with increasingly accurate high-dimensional data.  This process is called data assimilation and it is performed every day at all major operational weather centers across the world. Data assimilation  (DA) requires massive computing capabilities as realistic atmosphere-ocean models typically have billions of degrees of freedom. I will give a short overview of the ongoing research that aims to drastically decrease the required DA computational effort by reducing the dimension of the models involved and using stochastic perturbations to account for the unresolved scales. The incorporation of observation data is done by using particle approximations suitably adapted to solve high-dimensional problems.

We consider the problem of parameter estimation for a McKean stochastic differential equation, and the associated system of weakly interacting particles. The problem is motivated by many applications in areas such as neuroscience, social sciences (opinion dynamics, cooperative behaviours), financial mathematics, statistical physics. We will first survey some model properties related to propagation of chaos and ergodicity and then move on to discuss the problem of parameter estimation both in offline and on-line settings. In the on-line case, we propose an online estimator, which evolves according to a continuous-time stochastic gradient descent algorithm on the asymptotic log-likelihood of the interacting particle system. The talk will present our convergence results and then show some numerical results for two examples, a linear mean field model and a stochastic opinion dynamics model. This is joint work with Louis Sharrock, Panos Parpas and Greg Pavliotis. Preprint: https://arxiv.org/abs/2106.13751

Often in the context of data centric science and engineering applications, one endeavours to learn complex systems in order to make more informed predictions and high stakes decisions under uncertainty. Some key challenges which must be met in this context are robustness, generalizability, and interpretability.

We consider statistical inference for a class of agent-based SIS and SIR models. In these models, agents infect one another according to random contacts made over a social network, with an infection rate that depends on individual attributes. Infected agents might recover according to another random mechanism that also depends on individual attributes, and observations might involve occasional noisy measurements of the number of infected agents. Likelihood-based inference for such models presents various computational challenges. In this talk, I will present various sequential Monte Carlo algorithms to address these challenges.