Analyzing the Impact of Mini-Batch Size in Optimizing the performance of Deep Neural Networks

  • Subhadeep Guchhait, Dr. Smitha Rao2

Abstract

Deep neural network (DNN), has demonstrated impressive results in computer vision tasks including recognition of images, identification of objects, etc. But, the accuracy of a model depends on many parameters such as weights, bias, number of hidden layers, kinds of activation function, and hyperparameters. The correct tuning of hyperparameters directly affects the accuracy of a DNN model. In this paper, we will discuss two hyperparameters, one is the mini-batch size and the second one is nodes in each layer. It is also a daunting challenge to decide the optimum nodes number that are hidden in each layer and the input layer batch size. To get the most favorable number of nodes along with the batch size is very important in DNN as this effect directly on the training performance and the computational cost of our model. This paper summarizes the effect of these two hyperparameters in model building. Also, how to choose the optimal number of these two hyperparameters. Which has been determined by testing our model on five different datasets including CSV and image files?

 

Keywords: Deep Neural Network (DNN), Hyperparameters, Batch size, Hidden Nodes, classification, Architecture.

Published
2020-06-06
How to Cite
Subhadeep Guchhait, Dr. Smitha Rao2. (2020). Analyzing the Impact of Mini-Batch Size in Optimizing the performance of Deep Neural Networks . International Journal of Advanced Science and Technology, 29(05), 13074-13081. Retrieved from http://sersc.org/journals/index.php/IJAST/article/view/25906