Applying machine learning to information visualization

Abstract

The analysis of large amounts of complex data demands tools that facilitate knowledge extraction. Tools that provide visual representations of such data are highly attractive due to their simple and intuitive exploration and analysis mechanisms. User interaction with those visual representations is a central feature of such tools. Among existing techniques, sampling-based multidimensional projections have attracted great interest by enabling user knowledge incorporation into the creation of visual representations through interaction mechanisms with data samples. This kind of interaction could be modeled as a machine learning process. Despite recent work towards incorporating user knowledge in interactive visualizations, there are still limitations such as excessive user effort necessary to obtain satisfactory results and the lack of control over the transformations applied on the data being analyzed. This is the main objective of this master's project: model user interaction into a representation that could be used to create better visualizations, attempting to formalize current work as a learning proccess, understanding and defining the limits of such interaction. (AU)