| Description |
Professor Steffen Lauritzen (Oxford University)
Title : Gaussian Graphical Models
Slides :
Part1, Part2, Part3
Abstract
The lecture will be concerned with Gaussian graphical models and will go through the basic theory of Gaussian graphical models, models with additional symmetry properties, and alternative methods for estimation and model identification.
Professor Arnaud DOUCET (Oxford University)
Title : Sequential Monte Carlo methods for Bayesian Computation
Slides :
Part1, Part2
Abstract
The aim of this course is to introduce students to the class of Sequential Monte Carlo (SMC) methods and presents the most recent developments in this fast moving research area. These simulated-based methods have emerged over the past twenty years as an alternative and complementary approach to Markov chain Monte Carlo (MCMC) methods.
First lecture : I will introduce the main ideas behind SMC methods, discussing their properties and limitations. I will show how one can apply these techniques so as to perform inference in nonlinear non-Gaussian state-space models, Dirichlet process mixtures and changepoint models.
Second lecture: I will show how it is possible to use MCMC within SMC methods so as to obtain algorithms which are able to explore high-dimensional probability distributions and approximate normalizing constants. The resulting methodology will be demonstrated on various applications including non-standard filtering problems, inference in mixture models and approximate Bayesian computation.
Third lecture: I will show how it is possible to use SMC within MCMC methods to obtain a class of powerful MCMC algorithms to sample from high-dimensional target distributions and I will make the connection with pseudo-marginal algorithms. I will show how it is possible to use such techniques to perform Bayesian parameter inference in complex state-space models or design efficient algorithm to sample Dirichlet process mixture models.
Professore Di Cook (Iowa State University)
Title : Data Visualization: Discover, Explore and be Skeptical
Slides :
Part1, Part2, Part3
Abstract
Lecture 1: Exploratory data analysis
In this technological age we are drowning in data. Good data visualisation helps us to swim, digest the data, and learn about our world. The statistics community creates visualisation systems within the context of data analysis,
so the graphics are designed to support and enrich the statistical processes of data exploration, modeling, and inference. As a result, statistical data visualisation has some unique features which differentiates it from visualisations made in other fields. Statisticians are always concerned with variability in observations and error in measurements, both of which cause
uncertainty about conclusions drawn from data. Dealing with this uncertainty is at the heart of classical statistics, and statisticians have developed a huge body of inferential methods that help to quantify uncertainty.
Statistical data graphics cover a spectrum of methods including elegant static data visualisations and highly interactive and dynamic graphics used for exploratory data analysis.
In this lecture, we will explain how graphical methods were used to study the tech boom and bust of the late 1990s, food quality control, atmospheric CO2 levels and temperature changes, stimulus fund spending, university ranking, PISA education data, labor market wages, stock market trends, and soybean breeding for agribusiness. (Ok, only a selection of these will be used.)
Lecture 2: Plotting many dimensions
When we visualize data, we are interested in portraying abstract relationships among such variables: for example, the degree to which income increases with education, or whether certain astronomical measurements indicate grouping and therefore hint at new classes of celestial objects.
In contrast to this interest in abstract relationships, many other areas of visualization are principally concerned with the display of objects and phenomena in physical three-dimensional (3D) space, e.g., the display of humans in medicine or animated movies, manufacturing cars or animated movies, flow visualization in aeronautics or meteorology. In these areas one often strives for physical realism or the display of great detail in space, as in the visual display of a new car design or of a developing hurricane in a meteorological simulation.
The data visualization task is obviously different from drawing physical objects, and the challenge is to create pictures that reflect these abstract entities. One approach to drawing abstract variables is to create axes in space and map the variable values to locations on the axes, and then render the axes on a drawing surface. In effect, one codes non-spatial information using spatial attributes: position and distance on a page or computer screen. The goal of data visualization is translating abstract relationships to interpretable pictures. This way of thinking immediately brings up a limitation: Plotting surfaces such as paper or computer screens are merely two-dimensional (2D). In data visualization, however, the number of axes required to code variables can be large: Five or ten axes are common, but these days one often encounters dozens and even hundreds.
Overcoming the 2D and 3D barriers is a key challenge for data visualization, which can be met by using computer to mimic and amplify a strategy familiar from photography: taking pictures from multiple directions so the shape of
an object can be understood in its entirety: the ``multiple views'' paradigm. In our 3D world the paradigm works superbly, because the human eye is adept at inferring the true shape of an object from just a few directional views. Unfortunately, the same is often not true for views of abstract data. The chasm between different views of data, however, can be actively bridged with additional computer technology: to manipulate pictures, to pull and push their content in continuous motion like a moving video camera, or to query at objects in one picture and see them light up in other pictures. Motion links pictures in time; poking links them across space.
Lecture 3 Inference and statistical graphics
Plots of data often provoke the response ``is what we see really there''.
In this lecture we will discuss ways to give visual statistical methods an inferential framework. Statistical significance of ``graphical discoveries'' is measured by having the human viewer compare the plot of the real dataset with collections of plots of null datasets: plots take on the role of test statistics, and human cognition the role of statistical tests,
in a process modeled after the ``lineup'', popular from criminal legal procedures. This is a simple but rigorous protocol that provides valid inference, yielding p-values and estimates of the test power, for graphical findings. We will describe a series of experiments comparing the lineup protocol with classical tests, and how it can be used for quantifying
discoveries in data and for designing good data plots.
Pierre Jacob (Oxford University) - Workshop
Programme |
| 8h30-10h00 | | Arnaud Doucet | Pierre Jacob (Workshop)
| Di Cook | | 10h-10h30 | | pause café | pause café | pause café | | 10h30-12h00 | |
Steffen Lauritzen |
Pierre Jacob (Workshop)
| Steffen Lauritzen | | 12h00-14h00 | | pause de midi | pause de midi | pause de midi | | 14h00-14h45 | Thé de bienvenue | | | Steffen Lauritzen 14h00- 15h30 | | 15h00-16h30 | Di Cook | pause café | pause café | | | 17h00-18h30 | Arnaud Doucet | Di Cook | Arnaud Doucet
| | | 18h30-19h00 | apéritif | apéritif | Réunion Commission Scientifique | | | 19h15-22h00 | Repas du soir | Soirée raclette | Repas du soir |
| |
short URL