Weight Optimization Study of Partially Trained Artificial Neural Networks

Artificial Intelligence systems supporting human decision-making are becoming increasingly popular in many areas of entrepreneurship and business support. Fast, precise and accurate hint developed by an algorithm using an artificial neural network is often a guarantee of making an optimal decision and, ultimately, commercial success.

The purpose of this article is to present the concept of improving the performance of a partially trained Artificial Neural Network (ANN). The ANN training process, depending on the selected learning method and the used structure, can be quite long-lasting. Premature termination of the learning process may result in vague or ambiguous responses to ANN queries, or even their misinterpretation. The method presented in this article improves ANN performance without the need to continue the ANN training process. The method modifies ANN weights values, which has a direct impact on the obtained results and can be used to improve the performance of poorly trained ANNs as well as to enhance the operation of already trained ANNs.