Visualization of data structures and machine learning of rules
Yoh‐Han Pao · 2002
Intelligent task performing machines need to be able to benefit from experience and continue to improve its own task performance capabilities progressively. Towards that end, a machine capable of learning needs to be able to abstract bodies of complex high-dimensional data into manageable groupings and be able to discern and articulate relationships between such data items. This discussion describes how 2D depictions of multivariate data can support such machine learning activities. A new dimension-reduction procedure is described schematically. It seems to generate mappings which have useful 'topologically correct' characteristics. Previous related data abstraction schemes are discussed briefly for comparison purposes. In the case of interrelated tasks and corresponding bodies of 2D maps, the latter can facilitate the recognition of associations and the inference of rules, important components of machine learning.