Deep Learning

Could use a lot of work: not yet worth the money
A lot of effort has clearly gone into this course and I believe it may be worthwhile doing at some point, but unfortunately isn't yet worthwhile.

The materials are a mix-up of different components merged together and it shows. Some components are extremely lite on detail (they point you at blogs you should read and don't offer much more), while others repeat the same thing over and over (how many times do you need a picture of a mountain to explain SGD?).

The projects and assignments are a really unnatural format: you're given half-completed notebooks where you have a very incomplete picture of what's going on (because someone else wrote 80% of the code) and you're expected to just complete a few functions or variable assignments. This is useless for building up pragmatic knowledge that would be useful in the real world.

They've made an attempt at having "tutors" via a slack channel, a forum, and a mentor check-in system: these are fundamentally useless for the types of problems students are encountering. The slack is littered with fragments of code of students who can't get things working, and results in students apparently copy and pasting each-others code until something works. The community engagement team are active on the slack, but can't answer technical questions. The experts who drop in once a week for an hour or so generally aren't able to help, and it shows: they rarely get much interaction.

The course is very silicon valley in that it's "move fast and be broken": one of the most active channels on slack is for pointing out errors and problems in the code, materials, and videos.

That said, the production quality of the videos is generally quite high. The lecturers are all quite clear and well spoken, the videos have good diagrams and visuals most of the time.

Except for the inclusion of Siraj. seriously, who at Udacity signed off on including this guy as a teacher in this course: https://www.youtube.com/watch?v=XTNl5WxklgE

Ultimately, I do believe this course may at some point be worthwhile. Before then, save yourself $500 and buy:
- Deep Learning with Python by Francois Chollet (Author of Keras)
- Hands On Machine Learning with Scikit-Learn and Tensorflow by Aurelien Geron
- Deep Learning by Goodfellow et al.

BEST INTRO TO DEEP LEARNING
BEST INTRO FOR PEOPLE WITH LITTLE OR NO DEEP LEARNING EXPERIENCE

START
The best thing about the course is the first section. It takes a simple prediction problem, "If grades= something, age=something then will X get accepted into university". It applies "common" sense math to
it and after 15 mins you realize - "oh my god!" this is machine learning and better "Neural net". Excellent visuals, good explanation.

MIDDLE
Further sections are standard, good and challenges your understanding and will prove to be beneficial.
There is a section by Andrew Task - that will walk through NN code (by far best NN explanation ever!)

LAST
It wraps up by introducing advanced techniques , GAN, Reinforcement learning etc. but its standard and does not go deep :D (as it is meant for beginners and intermediate alike)

Good Intro for people with Little or no Deep Learning Experience
This ND starts by covering the mathematical foundations of Deep Learning, then moves you through some interesting types of Deep Learning networks and their applications.

The ND uses TensorFlow and Python, which are generally accepted in the field.

This is not a replacement for or a supplement to any kind of graduate-level work in Deep Learning, but at a few hundred dollars, it will at least get you to where you can hold a Deep Learning conversation and make some engineering contributions to work that's going on currently.

This ND does _not_ present a coherent conceptual overview of why the Deep Learning architecture works on the tasks it's good at, mainly because there isn't a good explanation at present; it works, so we use it.