Abstract
Conference Title: 2018 International Conference on Advanced Systems and Electric Technologies (IC_ASET) Conference Start Date: 2018, March 22 Conference End Date: 2018, March 25 Conference Location: Hammamet, Tunisia CNNs (Convolutional Neural Networks) achieves considerable performance on the practically important text tasks (e.g. text classification). Nevertheless, neural network's accuracy significantly deteriorates for complex applications. However, they require researchers to specify the exact architecture and set hyper-parameters, regularization parameters including the filter region size, and so on. For instance, all the parameters and randomly initialized values determine how accurate the model is, and the final result produced by the model. To tackle these challenges, many researches proposed good ideas. In this paper, we propose the use of Fish School Search (FSS) in Convolutional Neural Networks (CNNs) such models. The FSS is a modern and new algorithm for optimization purposes. It has succeeded to join the family of other optimization methods. The use of FSS on the training process aims to optimize the results of the solution vectors on CNN in order to improve text mining tasks. In our work, we propose a generic method that could be used and applied on text mining tasks as well as in several data science tasks.