Improving Hopfield Neural Network Performance and Parameters Investigation

 In this work, an appropriate setting of the Hopfield Network (HN) parameter was investigated and applied to the classical traveling salesman problem. The investigation on the requirement of raw data normalization was also carried out. Moreover, a modified training algorithm by embedded a heuristic called elitism for improving the performance of the conventional HN was additionally proposed. Computer experiments were implemented using various problem sizes. The results obtained from the experiments indicated that the appropriate setting of HN parameter should be specified with a low value. It was also found that the usage of raw data or normalized data did not influence on the performance of HN. Another experimental result suggested that the proposed hybrid HN did not only outperform conventional HN in terms of the equality of the results, but execution time was also faster.

 Keywords: Neural network, Meta-heuristics, Artificial intelligence, Traveling salesman, Combinatorial optimization.

 Corresponding author: E-mail: Pupongp@yahoo.com