How to start with Deep Learning Without Experience in Machine Learning
A Guide to quickly immerse yourself in Deep Learning.
Wait, what? Deep learning without machine learning? It's ridiculous. It is not possible. Stop right there!
I have been in the field of Deep learning over 2 years now, which is not enough, I know. However, I recently found good reasons and a possible way to start with deep learning — if you are really interested in working in the most advanced fields like Computer Vision, Natural Language Processing, Robotics.
Before we dive deeper into that, let's first understand the interrelation between machine learning, deep learning and artificial intelligence.
Machine Learning, Deep Learning
Deep learning is a part of Machine learning or rather to say deep learning uses a particular algorithm of machine learning called “Neural Network”. This algorithm was inspired by the human brain.
Now, where to start?
- You need to know basic python(Its a must).
If you don't know the basic then read, Learn Python 3 the Hard Way. It's more than enough.
- Then start out with Grokking Deep Learning book. This book should be the first step to start into deep learning if whether you have or not any background in machine learning. Once done, you will know the in-depth working of the different algorithm using Numpy from scratch, which is great! Along with that, you will come to know the different mathematical concepts (though not in-depth) as you go through.
- Then you can move into learning different mathematical concepts working behind it. For that, I will suggest reading Chapter 1 of the Deep Learning Book written by Ian Goodfellow which will be enough to know all the mathematics behind in the working and get to know the more important machine learning algorithms.
Note: It is never possible to learn all the algorithms at once. You will get to know them over time depending on your field of application.
For eg.: Q-learning is used in Reinforcement learning whereas CNN is used in Computer Vision.
However, if you know the inner working of the algorithms then you will get to know that all these algorithms are tuned according to their field of application.
4. If you desire to be a Deep learning researcher in the future, learning mathematics is very important. But just don't get started blindly as you will not understand what to learn and where to start and very soon you will lose your interest.
Rather start building simple projects from scratch without using any framework in that way you will get to know the inner mathematical working of different algorithms. If you get stuck in any mathematical concept refer to ‘3 blue 1 Brown’ videos or ‘Khan Academy’. Try to do with as many simple algorithms as you can. Then when moving out try to do the more complex algorithms using frameworks like PyTorch or TensorFlow as doing it from scratch will become very difficult and it's unnecessary.
Now when choosing Frameworks go for PyTorch or Tensorflow. However, I will suggest going with Pytorch as it would be easy while coming from a Numpy background and you will find yourself at home. However, TensorFlow is an excellent Framework as it provides more functionality with its latest introduction of version 2.0. To start with you can use the official document provided by them. It's a great resource!!
And finally, once you have completed the above in the pattern I have suggested, feel free to move around and explore the different fields related to deep learning. Then according to its requirement, keep on learning and try to read research papers and implement them on your own. In this way, you will know how to create a new model and you become a Deep Learning researcher.
Thank your for reading. Please share your comments and feedback on it.