Visiting Scholar IPCV: Master class -- Bayesian computation -- Marcelo Pereyra, Heriot-Watt University -- 09-10-11 décembre 2019du 09/12 and 11/12
"Cutting-edge Bayesian computation approaches for performing statistical inference and learning in very high-dimensional imaging inverse problems for computer vision".
Elle sont largement ouvertes à tous et auront lieu au bâtiment A21:
-- Lundi 09/12/2019, 10h-12h, salle A21-401,
-- Mardi 10/12/2019, 10h-12h, salle A21-401, et
-- Mercredi 11/12/2019, 10h-12h, salle A21-401.
Contact: Jean-François Giovannelli
Master classes on cutting-edge Bayesian computation approaches for statistical inference in very high-dimensional imaging inverse problems for computer vision. These master classes will focus on the following state-of-the-art topics.
1 - High-dimensional Markov chain Monte Carlo methods that are derived from stochastic differential equations, particularly methods based on Langevin and Hamiltonian dynamics.
2 - Optimisation-empowered proximal Markov chain Monte-Carlo methods that use elements of convex calculus and proximal optimisation, as well as stochastic optimisation algorithms that are driven by Monte Carlo methods, and their application to empirical Bayesian estimation for imaging problems.
3 - Deterministic Bayesian computation methods that tightly integrate convex optimisation and high-dimensional probability theory (for example, to construct confidence intervals and hypothesis tests).
The master class will have the following structure. Each will start by introducing the rationale and motivation for the methodologies in the context of image processing and computer vision, followed by a detailed description of the methods, their theoretical convergence properties, and efficient algorithmic implementation. This then followed by a Matlab demonstration on a challenging imaging inverse problem related to medical imaging or remote sensing (all demonstration codes will be made available to the students), and a conclusion.
Additionally, the demonstrations will also be used to introduce advanced Bayesian analysis concepts such Bayesian model selection in imaging, Bayesian uncertainty quantification, and estimation of model parameters (e.g., regularisation parameters). This will provide the students with an acute critical awareness of state-of-the-art Bayesian computation and analysis techniques for large-scale inverse problems and their application to image processing and computer vision and will significantly contribute to developing the students’ critical thinking skill.