When to Go Deep

Deep learning is exciting, so people want to use it for everything. That's a costly mistake. A professional knows when it's the right tool. And knowing when it isn't will save you weeks of work.

• Exciting, so overused
• A costly mistake
• Choose the right tool

Deep learning or not?

Here's a simple guide. For images, audio, video, and language, deep learning is unbeatable, because the raw data has structure that it can learn by itself. For ordinary business tables, gradient boosting still usually wins, and it's cheaper. If you have very little data, a pretrained network, or a simple model, is best. And if you need to explain each decision, start with something simpler.

What you need for this course

You need just a few things. Basic Python. A little comfort with the idea of a function. And you don't need an expensive computer. Everything here runs on a normal laptop, and for bigger jobs, free cloud notebooks give you a G P U. Our Machine Learning course is a helpful start, but it isn't required. We'll build all the maths that we need as we go.

What you will build

Here's our mission. You've just joined a small lab called Neuron Labs, and your job is to build Aria, an assistant that can see and read. Along the way, you'll build a digit reader, an image classifier, a language model that writes text, a transformer from scratch, and even a model that dreams up new images. Every piece is real, and every line runs on screen.

Summary