This paper proposes Gaussian-PSO-based structural learning and fuzzy reasoning to optimize the weights and the structure of the Feed Forward Neural Network. The Neural Network is widely used for various applications; though it still has disadvantages such as learning capability and slow convergence. Back Propagation, the most used learning algorithm, has several difficulties such as the necessity for a priori specification of the network structure and sensibility to parameter settings. Recently, research studies have introduced evolutionary algorithms into the learning to improve its performance. The PSO is a population-based algorithm that has the advantage of faster convergence. However, the total number of the weights in the Neural Network determines the size of each particle, therefore the size of the network structure is computationally time consuming. The proposed method improves the learning and removes the stress by eliminating the necessity of determining a detailed network.
ASJC Scopus subject areas
- コンピュータ サイエンスの応用