Coursera Deep Learning vs Fast.ai Deep Learning Course
I took deep learning course from Andrew NG first as a beginner and just now completed part 1 of fast.ai course. I want to list out few pros and cons of each course.
Coursera Deep Learning Course
It is a specialization with 5 courses dedicated to different field/facets of deep learning.
- It is very well structured like a book
- Andrew NG is a seasoned instructor and goes patiently through each concept
- It has in built quizzes and programming exercises. Although, it is possible to go through them without understanding the inner working, it at least forces you to spend some time in writing working code
- It has very good explanation of back-propagation
- There is a complete course on structuring your project. It covers practical advises on what to do when you get stuck while working with incomplete data, small dataset, low accuracy and other such issues
- You get a verified certificate which you can flaunt on SM
- Although, NG is obviously quite knowledgeable on how AI is used in the industry, apart from some anecdotes most concepts are covered with academic bent
- Course sometimes feel too slow and spends way too long to get us to a level at which average AI practitioner works
Fast.ai course Part 1
Fast.ai course is a series of lectures given by Jeremy Howard, a long time AI practitioner. It is completely free and open course.
- This course feels bit informal compared to Coursera course and is arranged in quite odd way and is loosely structured.
- Jeremy instead of going through each concept patiently takes you right to the working code which you are likely to write
- As Jeremy says, instead of constructing from basic things, the course starts with working cat/dog classifier and takes you into depth of everything from time to time
- The course is full of practical advice and anecdotes which makes you comfortable getting your own hands dirty
- Since there are no restrictions on access, you are not forced to do anything apart from watching the videos. You can easily get by just watching the videos and not writing a single line of code to your own detriment.
- Many basic concepts are glossed over (relative to Coursera course) rather than going deep in them
- Most of the coding examples are in Fast.ai library which is custom library built on top of Pytorch. Although, it performs very well, it has lot of magic sauce built into it. I am not sure how widespread it is but is not probably as widespread as keras.
- There is no way to verify you took the course or not.
To sum up, Coursera feels more like academic setting while fast.ai feels more like industry/practitioner setting. If you are starting with deep learning, you can start with either of those courses but each course will leave some gaps. Taking Coursera course first and then fast.ai course would likely be ideal.
Honestly you can go through each course at brisk pace and not worry too much about every minor detail. You would likely have to go back to them anyway and actual learning will come from doing your own exercises.