Intuitive Deep Learning: Introduction

Source: Deep Learning on Medium

Are you interested in knowing how Deep Learning works?

There are many excellent courses out there which offer a comprehensive technical introduction to Deep Learning. I’ve taken these courses personally — Stanford’s CS231N class, Andrew Ng’s Machine Learning and Deep Learning series on Coursera, just to name a few. The focus of these courses, however, is to train technical coders who perhaps one day might pursue a career in Machine Learning and Artificial Intelligence.

For people who do not intend to go down the technical road, they are often consigned to ‘high-level overview’ talks on AI. These talks hype up AI and its applications, treating everything behind the scenes as some magical ‘black-box’. They feature general themes, such as “This is how AI can transform your business…”, “AI is dangerous to society because it’s not explainable…”, and so on. In fact, you’ve probably watched one or two of them yourself! But while these talks might excite (or terrify) you about applying AI in your industry, taking the next step probably requires a deeper understanding of what lies within this magical black box. What can I really do with AI in the context of my business? What do people really mean when they say AI is not explainable?

There seems to almost be some sort of dichotomy — either give people a complete treatment of Deep Learning (with math and code included), or just give them a high-level talk on how exciting Deep Learning and forget about all the inner workings of Deep Learning.

But it doesn’t have to be that way! So I’m on a simple mission: take the middle ground and explain the intuition behind many cutting-edge deep-learning concepts to a fair degree of detail, all the while without going too much into the math or code behind it.

If you are a complete beginner and your aim is just to understand how Deep Learning works and what lies within this magical black-box, then this series is for you. My promise to you is that by the end of the series, you’ll understand deep learning concepts intuitively, even if you don’t know the math or the code behind them. You will have a deeper appreciation of the nuances behind Deep Learning algorithms, and you will be able to communicate with Machine Learning engineers using the same language and jargon, perhaps even debate and discuss with them some of the latest developments in the field.

And if your aim is just to understand Deep Learning concepts at purely the intuitive level, then this series is all you need and you can stop here.

If, on the other hand, you want to get your hands dirty into the math and the code and you’ve taken up a more comprehensive course on Deep Learning, then this series will be a good accompaniment for you. Many times, the manner in which we learn looks a little like this: wrangle with the math and code for many hours and the intuition of what the code is trying to do will somehow come to us eventually. We get thrown into the deep end of the pool, and if we miraculously manage to survive there then we’ll make our way to the shallow end of the pool.

I think that’s a pretty painful way of learning. A less painful way might be to really understand the intuition behind the concepts, focus on the ‘why’ and the bigger picture, and once that’s accomplished, the math and code will naturally make more sense. So I hope you take this series as that gentle introduction into that Deep Learning pool, a place to master your fundamentals and get a solid intuition and big-picture understanding before you go wading deeper and deeper.

And really, this isn’t limited to learning introductory Deep Learning concepts. Many state-of-the-art research papers are awfully difficult to read if you don’t understand the intuition behind what the researchers are trying to do! My aim is that this series will expand beyond the introductory topics, and cover the intuition behind these cutting-edge papers, helping researchers, industry professionals and graduate students alike to understand the papers better (or at the very least, provide a less painful way to understand the papers).

Well, I hope you’re excited for what’s to come!

This introductory part of this series is divided into four parts:

Part 1: Introduction to Neural Networks

  • Part 1a covers very high-level intuition of the Machine Learning paradigm
  • Part 1b covers some nitty-gritty details in making neural networks work

Part 2: CNNs for Computer Vision

Part 3: RNNs for Natural Language Processing

Part 4: Deep Reinforcement Learning

We cover very introductory material in these posts, well-suited for beginners in Deep Learning. There will be other posts beyond this introductory series where we cover more cutting-edge Deep Learning Concepts. I invite you to write on the topic that you’re most passionate about, and share the intuition with the world through this publication!

I sincerely hope you enjoy this series, and feel free to give me as much feedback as you can! It’s my first time writing like this, so there’s always room for improvements (: