Tek Gizli Katmanlı İleri Beslemeli Sinir Ağlarında Gizli Nöronların Etkisi

Abstract

Especially in the last decade, Artificial Intelligence is gaining popularity increasingly since deep learning and neural networks have fast and powerful machine learning-based techniques that can solve many real-time problems efficiently. In this study, Extreme Learning Machine (ELM), capable of high and fast learning is used for optimization parameters of Single hidden Layer Feedforward Neural networks (SLFN)s. The learning capability of such systems is directly related to the effectiveness of the parameters and the calculation methods. Hidden neurons number, one of the parameters in the calculations is discussed and its role is examined. The importance of the appropriate selection of this value will not only be emphasized but also a new method will be proposed for proper selection. The proposed method, Normalized Average Value (NAV) is a simple and effective formulation that originates from statistical methods in this field. Experimental results for determining a correct number of hidden neurons (L) show that random selection of this number causes either overfitting or under fitting problems. NAV can improve any algorithm in order to reach better learning rates. The results show that it provides a 10-15% performance increase due to random selection if the number of hidden neurons, L is determined according to the result of the study.