6 Deep learning for fraud detection

 

This chapter covers

  • Deep learning overview
  • The deep learning advantage over classical machine learning
  • Different types of deep learning models

You might think the person in figure 6.1 is real. Not really. This person does not exist! This is an imaginary person rendered by a deep learning model. The model, to be precise, is a StyleGAN2 model, short for Style Generative Adversarial Networks 2, which is one of the thousands of immensely powerful deep learning models that are changing the (digital) world as we know it. You can check out more such real-looking imaginary people at https://thispersondoesnotexist.com/.

Figure 6.1 A real-looking fictitious person’s face generated by a deep learning model.

With great power, comes great responsibility, and this couldn’t be truer for deep learning models. While you and I plan to use these models creatively to better humanity, fraudsters see these models as a massive upgrade to their fraud toolkit. Fraudsters can make up a non-existent identity using an imaginary face, as shown in the form of a New York driver’s license in figure 6.2, created using https://verif.tools/en/.

Figure 6.2 A real-looking fake identity document – New York driver’s license. The left image shows the rear side of the document and the right side shows the front side of the document.

6.1 What is deep learning?

6.1.1 Deep neural networks

6.1.2 What led to the resurgence of DL?

6.2 Deep learning versus classical machine learning

6.2.1 Implicit feature engineering in deep learning

6.2.2 Reviewing deep learning frameworks

6.2.3 Model selection, training, evaluation, and deployment in deep learning

6.3 Types of deep learning models

6.3.1 Understanding convolutional neural networks

6.3.2 Understanding recurrent neural networks

6.3.3 Understanding graph neural networks

6.3.4 Understanding transformers

6.4 Summary

sitemap