Title: Classification algorithm based on random iterated projections
Authors: Qi Xiao - Iowa State University (United States)
Zhengdao Wang - Iowa State University (United States) [presenting]
Abstract: Random projections are useful in reducing the dimensionality of a data set while almost preserving the distances between the data points. For this reason, they have found uses in many dimensionality-reduction applications. Algorithms have been proposed that use random projections for classification problems. In such algorithms, the projections are usually applied to the data points in one batch to map the data points to a lower dimensional space. We will present an algorithm that performs that projection in a sequential manner. The subsequent projection vectors, although dependent on the vectors used in earlier projections, are not required to be orthogonal to them. We will present the rationale for using such sequential projection, and also present numerical results that compare the proposed algorithm with existing methods.