Exploiting “Mental” Images in Artificial Neural Network Computation

Abstract

In Artificial Neural Network (ANN) computing the learned knowledge about a problem domain is “implicitly” used by ANN-based system to carry on Machine Learning, Pattern Recognition and Reasoning in several application domains. In this work, by adopting a Weightless Neural Network (WNN) model of computation called DRASiW, we show how the knowledge of a problem, internally stored in a data representation called “Mental” Image (MI), can be made “explicit” both to perform additional and useful tasks in the same domain, and to better tune and adapt WNN behavior in order to improve its performance in the target domain. In this paper, three case studies of MI processing in the realm of WNN applications are discussed with the aim of proving the viability and the potentialities of exploiting internal knowledge of WNNs to self-adapt and improve their performance.