Best way to learn DL from scratch in 2018
There are more and more amazing resources that make Deep Learning more accessible than ever. A few years ago, it would be extremely hard to find a good introduction that doesn’t overwhelm you with a gigantic list of prerequisites. Now, you don’t need eleven PhD’s to get started. If you know python — you can start training your first ANNs in a matter of days, and get quite comfortable in this field in a few months.
This is a collection of the best resources that will help you get started, that assumes no prerequisites other than basic understanding of Python.
Python for Data Science and Machine Learning
If you are unsure about your python skills, I recommend getting started here. This simple introductory course will take you through all the libraries you will need to comfortably follow any of the courses below. Use it to refresh your memory, and learn the fundamentals of Python, Jupyter, NumPy, and Matplotlib. There are more sections in this course, but as long as you know these 4 — you’re good.
3blue1brown videos on neural networks
The best way to understand the basics of DL is to watch this extremely brilliant high-level concise overview of how ANNs work. I highly recommend you get started here. Watch these 4 short videos, and you have a good general idea of everything you’ll need to learn in more detail using the other resources.
Grokking Deep Learning
This is a fantastic book that will teach you the fundamental principles of Deep Learning. It has very intuitive and easy to follow explanations, and doesn’t use any libraries other than NumPy. You will gradually build up your understanding, and build all the basic algorithms you want to know, from scratch.
Deep Learning With Python
Unlike the previous book, this one doesn’t go into math and theory, instead guides you through building several practical projects with a very simple to use DL library(keras). It’s a great way to gain practical experience in addition to the theory from the previous book. Also has no prerequisites other than python, and makes it very easy to get started.
Fast.ai Deep Learning course (v.2)
If you’re looking for the fastest, easiest way to go from zero to training cutting edge deep learning algorithms, the best place to learn is fast.ai.
This is a free course that teaches you deep learning using a top-down approach. You will begin by training a state of the art image classifier using only a few lines of code in the first couple of hours, and as you follow along, the course will go more in depth explaining the theory, other algorithms, and giving you a complete understanding of how things work.
The amount of information might be a little overwhelming at first, that’s why I recommend you to also use the books mentioned above if you get confused at any point.
These 5 resources will be enough to take you from zero to competently training ANNs in a few months. The rest of the article is a collection of other excellent resources that will make your journey easier.
The easiest way to learn the math you’ll need for DL is by following Khan Academy and 3blue1brown’s courses. You won’t need to learn all of this before other courses, but I recommend watching these videos in your free time, eventually you will find this knowledge useful. Most importantly, you will need to know these 3 subjects:
- Linear Algebra
- Probability and Statistics
Hands-On Machine Learning
This book is the best way to learn the machine learning in general, it gives you an excellent reference and overview of all the algorithms.
Andrew Ng’s Coursera course
This course if one of the most popular ways to get started with ML.
Once you’re ready for something more advanced, I highly recommend checking out these new brilliant courses from MIT:
- 6.S191 Introduction to Deep Learning
- MIT 6.S099: Artificial General Intelligence
- MIT 6.S094: Deep Learning for Self-Driving Cars
Finally, these are resources for people interested in learning about AGI and gaining a broader understanding of AI as a field.
The Master Algorithm
This is a great audiobook giving you a high-level overview of ML field and algorithms, aiming to lay down the roadmap towards the “Master Algorithm”(AI that can learn anything).
In this audiobook you will learn a theory of how human mind works that DL algorithms are based on. It’s very engaging, fascinating and fun to listen to, and will help you to think about ANN’s in a new way.
Artificial Intelligence: A Modern Approach
The leading textbook in Artificial Intelligence. It’s not the fastest way to get started, but it’s considered one of the best AI textbooks ever written.
Stanford AI course
Brilliant course based on Artificial Intelligence: A Modern Approach. Will really help you to understand this book better, and build a bunch of practical projects as you follow along.
Finally, a few great playlists on DL, just to complete the collection:
Source: Deep Learning on Medium