The impact of dataset complexity on transfer learning over convolutional neural networks

Abstract

This paper makes use of diverse domains’ datasets to analyze the impact of image complexity and diversity on the task of transfer learning in deep neural networks. As the availability of labels and quality instances for several domains are still scarce, it is imperative to use the knowledge acquired from similar problems to improve classifier performance by transferring the learned parameters. We performed a statistical analysis through several experiments in which the convolutional neural networks (LeNet-5, AlexNet, VGG-11 and VGG-16) were trained and transferred to different target tasks layer by layer. We show that when working with complex low-quality images and small datasets, fine-tuning the transferred features learned from a low complexity source dataset gives the best results.