Deep Learning in Banking

Chapter 2Image Processing and Convolutional Neural Networks

Summary

After reading this chapter, and going through the exercises, you will be able to:

• Understand the role that images play in risk evaluation.
• Understand the theory and main components behind Convolutional Neural Networks.
• Develop a classification model with images using both transfer learning and one from scratch, and evaluate them using Python.

It may seem odd to start this book by discussing images in banking. Images are, along with video, probably one of the least used sources of alternative data in modern banking. There is a simple reason why we may want to start by studying its use: image processing is the foundation of modern deep learning. Understanding how a deep learning model is structured for image processing and how the field has evolved will allow us to understand the more complex methodologies in subsequent chapters. It will also give us an opportunity to discuss some of the trickier aspects of the safe and trustworthy use of unstructured data in the field, particularly in credit risk.

From a technical point of view, this chapter will focus mainly on Convolutional Neural Networks (CNNs or ConvNets, for short). These models are excellent at creating new variables that summarize what segments of the input are relevant to the problem at hand, in particular, within data that can be expressed as a tensor, or a set of matrices. Images, video, audio, graphs, and most sources of unstructured data fit ...