An Epsilon-Greedy Mutation Operator Based on Prior Knowledge for GA Convergence and Accuracy Improvement: An Application to Networks Inference

This paper introduces a new mutation operator for networks inference based on the epsilon-greedy strategy. Given some prior knowledge, either provided by a third party method or collected from literature, our approach performs mutations by randomly exploring the search space with epsilon-frequency and by exploiting the available prior knowledge in the remaining cases. The algorithm starts with a highly exploitative profile and gradually decreases the probability of employing prior knowledge in the mutation operator, thus reaching a trade-off between exploration and exploitation. Tests performed have shown that the proposed approach has great potential when compared to the traditional genetic algorithm: it not only outperforms the latter in terms of results accuracy, but also accelerates its convergence and allows user to control the evolvability speed by adjusting the rate with which the probability of using prior knowledge is decreased.