23rd EANN / EAAAI 2022, 17 - 20 June 2022, Greece

Efficient Approaches for Data Augmentation by Using Generative Adversarial Networks

Pretom Kumar Saha, Doina Logofatu

Abstract:

  In the present and future, data is the most valuable thing in the world. Therefore, it is now a challenge for everyone in every sector to work with data. Collecting data to predict or collecting data to analyze is a very valuable task. Moreover every new research, new machine learning method, and algorithms testing depends on a massive amount of different data. Furthermore, it is also a security issue for many fields to share actual data. It is always hard to find the perfect data set. It is not just about figuring out huge amounts of data. Many other data analysis processes need to be performed on that dataset to make it worthwhile. To overcome this problem, data augmentation is one of the suitable solutions. The idea behind data augmentation is to create a new dataset that depends on some existing dataset features. Generative Adversarial Networks (GANs) are a class of machine learning frameworks introduced by Ian Goodfellow in 2014. They are a game-like way to learn and generate new datasets. GANs have two parts, one is the generator, and the second is the discriminator. They play against each other to win the game. We will use our data set on the GAN model using some specific hyperparameter value and optimizer, which we will find out through our experiment. Finally, we will produce a CSV file with model-generated synthesis data and visualize the performance statistic of our model in the graph. This article will explain the different facts related to Data Augmentation, and GANs.  

*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.