Hypercircle inequality for partially-corrupted data

Abstract

In recent years, the problem of learning and methods for learning functions have received increasing attention in Machine Learning. This problem is motivated by several applications in which it is required to estimate a function representation from available data. Recently, an extension of hypercircle inequality to data error (Hide) was proposed by Kannika Khompurngson and Charles A. Micchelli and the results on this subject have constructed a new learning method. Unfortunately, the material on Hide only applies to circumstances for which all data are known within error. In this paper, our purpose is to extend the hypercircle inequality to circumstances for which data set contains both accurate and inaccurate data.