Open AccessSearch-based approach to multilayer perceptron training
Author(s) -
Mirosław Kordos
Publication year - 2005
Publication title -
studia informatica system and information technology
Language(s) - English
Resource type - Journals
ISSN - 1731-2264
DOI - 10.5072/si2005_v26.n1.595
Subject(s) - perceptron , computer science , training (meteorology) , simplicity , algorithm , artificial neural network , variable (mathematics) , search algorithm , artificial intelligence , training set , machine learning , pattern recognition (psychology) , data mining , mathematics , meteorology , epistemology , physics , philosophy , mathematical analysis
The paper presents two search-based algorithms for MLP training; numerical gradient and variable step search algorithm. The advantages of the methods comparing to analytical gradient-based algorithms include low memory requirements, the algorithm simplicity and determining more optimal next step direction by direct access to the influence of hidden layer weights on network error.
Accelerating Research
Robert Robinson Avenue,
Oxford Science Park, Oxford
OX4 4GP, United Kingdom
Address
John Eccles HouseRobert Robinson Avenue,
Oxford Science Park, Oxford
OX4 4GP, United Kingdom