A Great Learning Stratagem through Artificial Bee Colony Technique for Grouping

Abstract

There have been many attempts to create training algorithms that can successfully train neural networks in a reasonable amount of time while also achieving acceptable generalisation performance since the inception of neural networks itself. The training of feed-forward neural networks using outmoded gradient grounded learning procedures, such as BP (Back Propagation), often gets trapped in local minima and requires a great deal of time to converge. All of the network weights and biases must be tweaked iteratively for these algorithms to perform properly. Recently, a new training method for randomised feed-forward neural networks known as the ELM (Extreme Learning Machine) has gained prominence. With ELM, the output weights are computed analytically, while the hidden layer parameters are first seeded at random and held constant throughout the learning process. Over the past decade, the ELM framework's popularity has skyrocketed due to its quick training speed and wide applicability. This study delves further into the subject of ELM and the research questions that have arisen around it. Methods based on swarm intelligence are articulated clearly. ELM's recent controversy is also discussed at length. Then, a fresh ELM K-means approach is created. The K-means approach is used to accomplish clustering in the high- dimensional feature space that has been projected from the data using ELM. Then, a hybrid strategy, ELM-ABC, is described, which employs the ABC algorithm to carry out clustering in the ELM topographic space. The method is an improvement over the ELM K-means approach.