Alternating Implementation of Machine Learning Assisted and General Hill-Climbing Algorithm

Combining machine learning techniques with meta-heuristic optimization algorithms is an emerging and exciting field for developing more effective optimization methods. This study focuses on enhancing the hill climbing algorithm by using a neural network to estimate the optimal step size in each iteration. This requires building a neural network architecture, generating training data, and training the model. This network is able to estimate the step size to be chosen during the search. Experimental tests have shown that this method alone is not reliable, it needs re-calibration steps after some iterations. This article deals with the frequency with which these steps should be incorporated into the implementation to achieve ideal performance.