Publications (1)0 Total impact
ABSTRACT: In this report a multi-dimensional data scaling approach is proposed in data mining and knowledge discovery applications. We derive the method based on an analogy to the physical computation of signal distortion. A dynamical cascade computation diagrams result from the statistical physics model computation in the free energy decomposition. We assess the scale invariance of various data sets, such as with the image motion sequences, and with the high dimensional chemical data sets. Theoretical model of error propagation is given by the numerical computational schemes. Statistical mapping of the data is analyzed through dynamical cascades, as a way of approaching its coding and control data structure. We show how it correlates by segmenting set of chemical compounds observations in a high dimensional property space. The proposed algorithm, also, is suitable for the implementation in parallel computer architectures. An example implementation on the multicore processors is given in the end of this report.
IU, report. 01/2007;