I³MS - Haiat Seminar
Dr. Gauillaume Haiat - Multiscale Modeling and Characterization of the Bone-Implant Interface
, Universitè Paris-Est, Paris
Implants are often employed in orthopaedic and dental surgeries. However, risks of failure, which are difficult to anticipate, are still experienced and may have dramatic consequences. Failures are due to degraded bone remodeling at the bone-implant interface, a multiscale phenomenon of an interdisciplinary nature which remains poorly understood. The objective of this seminar is to provide a better understanding of the multiscale and multitime mechanisms at work at the bone- implant interface. To do so the evolution of the biomechanical properties of bone tissue around an implant will be studied during the remodeling process. A methodology involving combined in vivo, in vitro and in silico approaches is proposed. Different aspects related to the biomechanical behaviour of the bone-implant interface the static (contact problems) and in the dynamic (acoustics) regimes will be tackled.
CHARLEMAGNE DISTINGUISHED LECTURE SERIES-LeCun Seminar
Prof. Dr. Yann LeCun - Unsupervised Learning: the Next Frontier in AI
, Facebook, USA
Facebook AI Research & New York University
The rapid progress of AI in the last few years are largely the result of advances in deep learning and neural nets, combined with the availability of large datasets and fast GPUs. We now have systems that can recognize images with an accuracy that rivals that of humans. This will lead to revolutions in several domains such as autonomous transportation and medical image analysis. But all of these systems currently use supervised learning in which the machine is trained with inputs labeled by humans. The challenge of the next several years is to let machines learn from raw, unlabeled data, such as video or text. This is known as predictive (or unsupervised) learning. Intelligent systems today do not possess "common sense", which humans and animals acquire by observing the world, by acting in it, and by understanding the physical constraints of it. I will argue that the ability of machines to learn predictive models of the world is a key component of that will enable significant progress in AI. The main technical difficulty is that the world is only partially predictable. A general formulation of unsupervised learning that deals with partial predictability will be presented. The formulation connects many well-known approaches to unsupervised learning, as well as new and exciting ones such as adversarial training.
I³MS - Gray Seminar
Prof. Dr. Nico Gray - Particle Size-Segregation and Rheology of Dense Granular Flows
, The University of Manchester, UK
Hazardous geophysical mass flows, such as snow avalanches, debris-flows and pyroclastic flows, often spontaneously develop large particle rich levees that channelize the flow and enhance their run-out. Large scale experiments with 10 cubic metres of water saturated sand and gravel flowing down the 80m USGS debris-flow flume indicate that a subtle segregation-mobility feedback effect is responsible for their formation. Within the flow large particles segregate to the faster moving near surface layers and are preferentially sheared towards the front. Here they may be over-run, re-segregated and recirculated, to create a coarse grained front that is more resistive to motion than the more mobile ?finer grained interior. As a result the large particles are shouldered to the side to create static levees that constrain the flow laterally. Simple models for particle segregation and the depth-averaged motion of granular avalanches are described and one of the first attempts is made to couple these two types of models together. This process proves to be non-trivial, yielding considerable complexity as well as pathologies that require additional physics to be included. Some of these difficulties can be overcome by incorporating a depth-averaged mu?(I)-rheology for granular flow into the model. However, the mu?(I)-rheology turns out to have regions of ill-posedness itself at high and low inertial numbers.
EU Regional School - McClarren Seminar
Prof. Dr. McClarren - Polynomial Chaos Expansions for Uncertainty Quantification
Texas A&M University
In computational science and engineering one often deals with computer simulations where inputs to the calculation are uncertain. A natural question to ask is how uncertain the output of a simulation is given uncertainties in the inputs. In this lecture I will give cover the application of orthogonal expansions in probability space (also known as polynomial chaos expansions) to determine the distribution of quantities of interest from a numerical simulation. I will detail how to apply these methods to a variety of input uncertainty distributions, and give concrete examples for simple functions as well as non-trivial applications. The examples will also be an opportunity to point out to students the pitfalls and common mistakes that can be made applying these techniques. Finally, I will cover more advanced ideas such as sparse quadrature and regularized regression techniques to estimate expansion coefficients.
I³MS - Helzel Seminar
Prof. Dr. Christiane Helzel
Starting point of our considerations is a coupled system consisting of a kinetic equation coupled to a macroscopic Navier-Stokes equation describing the motion of a suspension of rigid rods under the influence of gravity. A reciprocal coupling leads to the formation of clusters: The buoyancy force creates a macroscopic velocity gradient that causes the microscopic particles to align so that their sedimentation reinforces the formation of clusters of higher particle density. Since the coupled system is high-dimensional, we are interested in the derivation of simpler systems which describe the dynamics without resolving the kinetic equation. We discuss two different approaches to obtain such systems. Furthermore, we discuss the numerical methods which were used to approximate the different mathematical models and show numerical results. This is joint work with Athanasios E. Tzavaras from KAUST.