A Visual Cognitive Method Based on Hyper Surface for Data Understanding

摘要

Classification is a fundamental problem in data mining, which has extensive applications in information technology. Data understanding is highly relevant to how to sense and perceive them. However, the existing approaches for classification have been developed mainly based on dividing dataset space, less or no emphasis paid on simulating human or animal visual cognition. This chapter attempts to understand visual classification by using both psychophysical and machine learning techniques. A new Hyper Surface Classification method (HSC) has been studied since 2002. In HSC, a model of hyper surface is obtained by adaptively dividing the sample space and then the hyper surface is directly used to classify large database based on Jordan Curve Theorem in Topology. In this chapter we point out that HSC is a data understanding method which accords with visual cognitive mechanism. Simulation results show that the proposed method is effective on large test data with complex distribution and high density. In particular, we show that HSC can deal with high dimensional data and build corresponding visual hyper surface using dimension transposition or ensemble method which accords with visual dimension transposition and multi-dimension cognitive mechanism respectively.