Interactive tools for pattern discovery

Real-world pattern recognition problems contain rich context that must be taken into account in solution development. Beyond the core classification tasks, there are several common challenges including the need to correlate noncomparable groups of variables while preserving data proximity relationships within each group, visualization of data geometry, rapid trial of many methodological alternatives, and integration with existing infrastructure. We explore the effective methods and tools to meet with these challenges. We describe a graphical software, Mirage, as an experiment to address such concerns.