Accuracy comparison of different batch size for a supervised machine learning task with image classification

Abstract

Machine learning is a type of artificial intelligence where computers solve issues by considering examples of real-world data. Within machine learning, there are various types of techniques or tasks such as supervised, unsupervised, reinforcement, and many hyperparameters have to be tuned to have high accuracy especially in image classification. The batch size refers to the total number of images required to train a single reverse and forward pass. It is one of the most essential hyperparameters. In our paper, we have studied the supervised task with image classification by changing batch size with epoch. The characterization effect of increasing the batch size on training time and how this relationship varies with the training model have been studied, which leads to extremely large variation between them. According to our results, a larger batch size does not always result in high accuracy.