Alternative Approximation Method for Learning Multiple Feature

Kannika Khompurngson, Suthep Suantai

Abstract

A foundational concept in learning problem is to constructa functional representation from  given data.Among learning methods,  Hypercircle inequality (Hi)has been applied to kernel-based machine leaningwhen data is known exactly.Recently, we have extended   Hi to data error in two ways: First, we have extended it tocircumstance for which all data is known within error.Second, we have extended it to partially-corrupted data.That is, data set contains both accurate and inaccurate data.In this paper, we apply the material from both previous work to  estimatethe unknown vectors in Hilbert space from knowledgeof both its norm and linear observations of it, known within  error.