Title :
Finding error data for linear separable model
Author :
Ota, Yuki ; Tanaka, Masahiro
Author_Institution :
Dept. of Inf. Sci. & Syst. Eng., Konan Univ., Kobe, Japan
Abstract :
In this paper, we consider determination of separating hyperplane where some data points are treated as noise. Pocket algorithm with ratchet is an algorithm for this kind of problems, but it does not guarantee minimal number of noisy instances. By noticing that a hyperplane can be determined by selecting n points among data (where n is the dimension of the data), we can guarantee that point. Actually, we found that the second and the third category of iris data is linearly separable by admitting only one instance of noisy data.
Keywords :
data handling; multilayer perceptrons; noise; error data; linear separable model; multilayer perception; noisy data; pocket algorithm; separating hyperplane determination; Data engineering; Electronic mail; Informatics; Information science; Iris; Machine learning algorithms; Noise reduction; Systems engineering and theory; Training data; Working environment noise; Perceptron; linearly separable; noisy data;
Conference_Titel :
ICCAS-SICE, 2009
Conference_Location :
Fukuoka
Print_ISBN :
978-4-907764-34-0
Electronic_ISBN :
978-4-907764-33-3
