Matthew Ward - Mixing Models and Data
Speaker: Matthew Ward, Professor of Computer Science, Worcester Polytechnic Institute
Time: Fri 11/11 at 15:00-16
Place: VR theatre, Kopparhammaren
A common task in data analysis is to develop a computational model of the data, using tools from fields such as statistics, pattern recognition, and data mining; these models can be used for many tasks, such as concisely describing the characteristics of the data as well as predicting future behaviors of the process generating the data. The problem is that many different models can be used to describe a given dataset, including variations on the same class of models (e.g., by using different parameters) or very different types of models (e.g., clustering, linear classifiers, association rule mining, or regression models). The field of visualization has been primarily focused on the display of data to enable analysts to either develop or confirm models of data behavior. There are also a number of examples where computed models are superimposed on the data to ease this confirmation task. Mostly this is done one model at a time. Over the past two years we have been exploring the design of techniques for both visual exploration of model space as well as interactions between model space and data space. Initially this was focused on the visualization of the parameter space for a single type of model to identify regions of this space that fit the data particularly well. More recently
our focus has shifted to visualizing collections of heterogeneous models to identify features such as robust regions in model space as well as subspaces that have yet to be explored. In this talk I'll give an overview of this work, along with some demonstrations of our prototype tools. I will also present some ideas for our future research directions.
Last updated: Mon Aug 25 10:12:50 CEST 2014