Introduction

In this section, we first define and train a simple neural network from scratch. This will give us an idea of the ingredients involved and and what they’re good for. We then start using features from torch, incrementally replacing functional building blocks by their torch equivalents.

Depending on your background, this section may be a quick and easy read; or you might find you’d like to consult additional resources, such as the ones mentioned in the introduction. Whereever you are on that spectrum though, our focus here is to call intention to the main concepts / building blocks, so you can spot them quickly and either quickly add them to your “internal inventory,” or arrange for further background reading.

In this running example, the network we’ll build is a toy one, about the shallowest you could make a “deep net.” Don’t worry though – once we’ve covered the basics, we’re ready to dive into applications, which will range from introductory and pedagogical to real-world and relevant to various scientific and business areas.