April 2026
| Date: |
14th April 2026 |
| Location: | Amphi F107 - Centre INRIA Grenoble Alpes - Montbonnot-Saint-Martin |
| Time: | 15:00-16:00 |
Speakers :
This seminar will have two interesting presentations:-
Yiye Jiang, postdoc at STATIFY team (Inria) : A new sampling algorithm of Shrinkage Inverse Wishart
Website Slides
Sampling from a probability distribution is a classical question and is important in Bayesian inference. For example, an ancient sampling algorithm is the one of Bernoulli(0.5), which is given by the procedure of tossing a homogeneous coin. An sampling algorithm of Uniform(0,1) can be given by pointing in a line segment with eye closed. -
Emile Hohnadel, postdoc at ELAN team (Inria) : Efficient 2D rod model with frictional contact
Website
During my thesis, I focused on the study of fibre assemblies through numerical simulations. The focus is usually put on the quality of the geometry over the produced forces, especially in the computer graphics community. The study of forces is however important in many domains, haptic feedback or sound generation to name a few. In this talk, I present a model of two-dimensional fibres. It improves on the previous state of the art as shown through careful validation numerically, in geometry and in force. This validation also includes contact detection and dry frictional contact response. Finally, I present some complex scenarios that are easy to construct with a dedicated library named Circonflex.
In this talk, I will introduce the problem of sampling by proposing some potentially interesting examples to solve together. These examples go gradually from univariate discrete prabability mass to bivariate continuous probability density, with the complicated ones recyclying the simpler ones. Through these samples, the key ideas of constructing a sampling procedure for a complicated distribution are conveyed.
In the second part of the talk, I will present a sampling algorithm for a recently proposed distribution Shrinkage Inverse Wishart whose domain is all postive definite matrices, which relies essentially on the conveyed ideas.
✉ Contact us