Beginning Anomaly Detection Using

CHAPTER 3

Introduction to Deep 

Learning

In this chapter, you will learn about deep learning networks. You will also learn how 

deep neural networks work and how you can implement a deep learning neural 

networks using Keras and PyTorch.

In a nutshell, the following topics will be covered throughout this chapter:

•  What is deep learning?

•  Intro to Keras: A simple classifier model

•  Intro to PyTorch: A simple classifier model

 What Is Deep Learning?

Deep learning is a special subfield of machine learning that deals with different types of 

artificial neural networks. Drawing inspiration from the structure and functionality of 

a brain, artificial neural networks at their core are layers of interlinked, individual units 

call neurons that each perform a specific function given input data.

In “deep” learning specifically, some of the best models consist of dozens of layers 

and millions of neurons, and have been trained on multiple gigabytes of data. Generally, 

deep learning models don’t always need to be this big to perform well on certain tasks, 

and the tasks that the large models are expected to perform are complex, ranging from 

outlining a wide variety of objects within an image to generating summaries of articles.

Thanks to recent increases in the computational power and availability of GPUs 

(graphics processing units), anyone with access to a decent enough GPU can train their 

own deep learning models, keeping in mind that larger models might require more GPU 

resources such as memory.

---

74

Today, deep learning is taking the world by storm thanks to the extreme versatility 

and performance that it offers. More traditional models in machine learning have a 

problem where adding more training samples leads to a plateau in performance, but 

that problem doesn’t exist with deep learning. Instead, deep learning models get better 

and better with more samples, meaning they scale far better in terms of data set size 

and gain better performance as a result. Deep learning models can be applied to nearly 

any task with resounding success, and so are employed in the fields of cybersecurity, 

meteorology, finances and stock markets, speech recognition, medicine, search engines, 

etc. What exactly about deep learning makes it so great? First, let’s take a look at what an 

artificial neural network is.