Practical Deep Learning
Neural networks you can train, debug and justify
Created by Daniel Mercer
What you will learn
- Write and debug a training loop from scratch before reaching for abstractions
- Diagnose a model that is not learning, and tell that apart from one overfitting
- Apply convolutional and transformer architectures to the right problems
- Fine-tune a pretrained model on a small dataset without wrecking it
- Manage experiments so results are reproducible weeks later
- Judge whether deep learning is the right tool at all
Course content
14 sections · 194 lectures · 30h total length
- What this course covers, and what it does notPreview04:00
- How to get the most out of itPreview04:00
- Setting up and downloading the exercise files06:00
- Tensors and autogradPreview14:00
- Tensors and autograd in detail10:00
- A closer look at tensors and autograd05:00
- Worked example: tensors and autograd13:00
- Where tensors and autograd goes wrong07:00
- Exercise: tensors and autograd15:00
- Tensors and autograd in practice11:00
- Two approaches to tensors and autograd06:00
- Tensors and autograd — questions from previous students12:00
- Reviewing your work on tensors and autograd08:00
- Tensors and autograd — the edge cases10:00
- A second worked example: tensors and autograd14:00
- Applying tensors and autograd to your own project06:00
- When tensors and autograd is not enough11:00
- Taking tensors and autograd further15:00
- Recap: tensors and autograd07:00
Topics covered
Requirements
- Solid Python and comfort with numpy
- Machine learning fundamentals — validation, overfitting, metrics
- Access to a GPU is helpful; free options are covered
Description
Deep learning is unusually easy to do badly and hard to debug, because the failure is silent. This course builds up from a hand-written training loop so you can see every moving part, then covers the architectures and the fine-tuning workflows in current use.
Your instructor
ML engineer, ex-search infrastructure
Daniel spent years making search results better for millions of people, which mostly meant evaluation, not modelling. He teaches machine learning with the measurement work put back in.
Learner reviews
- 5 star73%
- 4 star22%
- 3 star3%
- 2 star1%
- 1 star1%
- PAPetra A.1 April 2026
I have recommended this to two colleagues already. The structure makes it easy to dip back into.
- BVBen V.10 January 2026
Clear explanations without the usual filler. I watched most of it at 1.25x and still followed everything.
- ARAisha R.6 February 2026
Exactly what I needed. It assumes you are an adult and does not over-explain.
- VVViktor V.30 May 2026
Concise and honest about what it does not cover, which I appreciated.
- GOGrace O.6 December 2025
Too much time on setup and not enough on the harder material I actually signed up for.
- NDNiall D.5 March 2026
Very practical. I applied something from module four at work the same week.
- GCGrace C.14 February 2026
I came in knowing a bit and still learned plenty. The middle third is the strongest part.
- TLTariq L.2 January 2026
Concise and honest about what it does not cover, which I appreciated.
- OAOlive A.8 March 2026
I have recommended this to two colleagues already. The structure makes it easy to dip back into.
Showing 9 of 1,364 reviews.
Learners also looked at
More from Data Science and beyond.
Modern JavaScript From the Ground Up
Amara Osei
