The pandemonium system of reflective agents

Abstract

The Pandemonium system of reflective MINOS agents solves problems by automatic dynamic modularization of the input space. The agents contain feedforward neural networks which adapt using the backpropagation algorithm. We demonstrate the performance of Pandemonium on various categories of problems. These include learning continuous functions with discontinuities, separating two spirals, learning the parity function, and optical character recognition. It is shown how strongly the advantages gained from using a modularization technique depend on the nature of the problem. The superiority of the Pandemonium method over a single net on the first two test categories is contrasted with its limited advantages for the second two categories. In the first case the system converges quicker with modularization and is seen to lead to simpler solutions. For the second case the problem is not significantly simplified through flat decomposition of the input space, although convergence is still quicker.

Variations of the two-spiral task

“…The concept of data set partitioning or divide and conquer was employed by a large number of neural network studies which included tests on the spirals data (Whitley and Karunanithi, 1991;Smieja, 1996;Chen et al, 1997;Lu and Ito, 1999;Fu et al, 2001;Castro et al, 2000;Chen and Chang, 2000;de Souza et al, 2002). The concept of partitioning can successfully be employed to solve the two-spiral task without using neural networks at all.…”

A Robust Evolutionary Algorithm for Training Neural Networks

“…In the neural network community, learning to tell two spirals apart is a benchmark task which is an extremely hard classification task [3,4,26,27]. The learning goal is to properly classify all the training data (97 points on each spiral, as shown in Fig.…”

Reflective controls for intelligent distributed objects

“…Actor model-based reflective system such as [2], [8] and [16], have the role of meta-object expanded to permit the management of request scheduling. However, these implementations lack the capability for expressing synchronization schemes involving passed information.…”

“…Reflective systems allow computations to examine and modify properties of their own behavior with consideration for information from the external environment [8,13,16,19]. These systems are examples of meta-level architectures; software architectures which operate not only on a base level of normal computation, but also on a meta-level of computation about the system's operation.…”