Practical Deep Learning for Coders

The teaching order

Most machine learning courses spend weeks on mathematics before you touch a model. Practical Deep Learning for Coders, taught by Jeremy Howard, does the opposite: you train a useful image classifier in lesson one, then progressively open up the pieces to see how they work. For people who already write software, learning by taking apart something that works is far more effective.

The course site says its videos have been viewed more than 6 million times, and lists alumni who went on to roles at Google Brain, OpenAI, Adobe, Amazon and Tesla.

What Part 1 covers

The current Part 1, recorded at the University of Queensland in 2022, has 9 lessons of roughly 90 minutes each. It covers:

• Computer vision, image classification and deployment of a model as a small web app.
• Natural language processing with Hugging Face Transformers.
• Tabular data, including random forests and regression, so you learn when deep learning is not the answer.
• Collaborative filtering for recommendation systems.
• How neural networks work under the hood, built up from simple foundations.

It uses PyTorch, the fastai library and Hugging Face libraries.

Part 2

A separate Part 2, Deep Learning Foundations to Stable Diffusion, has 17 lessons for more experienced students. It rebuilds much of the stack from scratch and works up to implementing diffusion models, so the earlier high-level tools are fully accounted for.

Before you start

The course states its prerequisites: about a year of coding experience, preferably in Python, and high school mathematics. No deep learning background is needed. It recommends free cloud notebooks such as Kaggle rather than your own computer for training, unless you are comfortable with Linux system administration.

How to get the most from it

• Watch each lesson, then reproduce the notebook with the video closed.
• Apply it to your own dataset after each lesson; the course insists on this.
• Read the matching chapter of the accompanying book, Deep Learning for Coders with fastai and PyTorch, which is free online as notebooks.
• Use the forums at forums.fast.ai, which are active and beginner-friendly.
• Plan about 10 hours a week, so Part 1 takes roughly two to three months done properly.

Where it falls short

The fastai library adds a high-level layer on top of PyTorch that many workplaces do not use directly. The concepts transfer; some specific APIs do not. The Part 1 recordings date from 2022, so for the newest large language model techniques you will need more recent material. The teaching style is fast and informal, which some learners find hard to follow without the notebooks.

Common questions

Is it really free? Yes, the videos, notebooks and book are free.

Do I need a GPU? No, use Kaggle or another free cloud notebook.

Is there a certificate? No, the course focuses on projects rather than credentials.

Best for: working developers and students who want to build real machine learning models quickly and learn the theory as they go.