Multi-layer Perceptron Training Using Hybridized Bat Algorithm

Neural networks (NNs) are a subset in the field of machine learning (ML) that tends to make it possible for a machine to learn and make new predictions based on previous experiences and provided data. It is important to emphasize that there is no need to program this kind of behavior since the whole process is supported by the “self-adjustment” of the algorithm, which can evaluate itself and therefore adjust its parameters to get better performance and accuracy. Neural networks are different from other types of machine learning algorithms in such a way that they do not use statistical and mathematical models to make future predictions. Instead, they replicate the structure and the processes that happen inside the human brain. However, this type of learning is very computationally expensive since there is an enormous amount of states and conditions in which the network itself can be found. Therefore, it can be said that a process of learning for neural networks relates to the collection of NP-complete problems because of the large search space of possible solutions. Swarm intelligence (SI) algorithms can help to reduce this large space of solutions by finding a solution that is not optimal but close optimal and provide satisfactory results given how much longer it would take to train a network without them. In this paper, the authors have proposed a solution for the stated problem based on hybridized bat algorithm.