Summary
In this paper, a many-objective training scheme for a multi-layer feed-forward neural network is studied. In this scheme, each training data set, or the average over sub-sets of the training data, provides a single objective. A recently proposed group of evolutionary many-objective optimization algorithms based on the NSGA-II algorithm have been examined with respect to the handling of such problem cases. A modified NSGA-II algorithm, using the norm of an individual as a secondary ranking assignment method, appeared to give the best results, even for a large number of objectives (up to 50 in this study). However, there was no notable increase in performance against the standard backpropagation algorithm, and a remarkable drop in performance for higher-dimensional feature spaces (dimension 30 in this study).
See the full content of this document
Extract
Many-Objective Training of a Multi-Layer Perceptron
1. Introduction
The optimization of a neural network function representation by evolutionary algorithms is a long-termed and successful research field. The main advantage is the much simpler specification of the optimization goal, with the reduced need to establish a complex training rule, e.g., for the application of a corresponding backpropagation algorithm - where the latter task can easily become highly intractable even for rather simple modifications of the structure of a neural network (especially by introducing fuzziness). While most of these Neuro-GA approaches [16] were focussing on the minimization of a single objective like the mean-squared error, recent advances in the evolutionary multi-objective optimization (EMO) field also allowed for the consideration of further objectives for the training by using an EMO algorithm. Some newer approaches focus on the multi-objective creation of neural networ...See the full content of this document
Sponsored links
