Hello! This is the first post in a series of where I will be documenting my journey of studying deep learning (DL) among other kinds of artificial intelligence (AI). In this post I will briefly describe what deep learning is and what it is used for. I will then give a quick rundown of what readers can expect the structure of the blog to look like.
One of the first things in the field of DL that really caught my eye was this website my dad sent me last year:
This website has a stock of many photographs of people's faces as well as many images of faces that it has generated. The people do not exist, the images are completely made up. Despite this, the game (i.e. determining which is real) is rather difficult for someone who has not studied the computer programming. Can you determine which face is real? I was shocked to discover that an artificially intelligent system had come so far, in such a uniquely human, quasi-sacred domain.
Another really cool introduction to DL applications is the Netflix documentary on AlphaGO, a program that beat the world champion of the classic strategy game "go" which has long been thought of as a game too intuitive for a computer to understand. I will eventually return to this topic in a future post as computers playing strategy games fascinate me.
I could ramble on about the cool things going on in the world of DL but I should probably save that for after you guys have been introduced to the basics... So let's get our feet wet!
What Is Artificial Intelligence?
AI is the largest category, including any computer program that can make decisions. This can be as simple as a bunch of "if" statements which are the most basic form of logic gate in coding. AI is so broad that it's hardly a useful term. It includes deep learning algorithms such as the face generator we saw above, but it also includes tax softwares.
Unless specifically stated, I will not use AI to refer to a futuristic doomsday device known as "General AI" (although, I may do a post on that, just for fun). To quote my good friend Madhav Singhal:
"If one wants to philosophize about AI, then one should be able write the code and do the math for it first."
I agree. So that will be my first focus for this blog series.
What is Machine Learning?
Another term you have probably heard is machine learning (ML). ML is a subset of AI algorithms that has the ability to change itself so that it "learns" as it is exposed to more data. ML uses "artificial neural networks" to learn and make statistical predictions.
What is Deep learning?
As you may have noticed from the title, this blog will be focus on "deep learning". Deep learning has been responsible for the boom in popularity of AI over the past decade. Deep learning specifically refers to machine learning using an artifical neural network that has more "layers" (hence "deep") meaning that it can learn more complex patterns.
AI, even including DL, has been described as "Narrow AI". That is to say, it performs very well but only at a narrow task. For example, recognizing bone fractures or classifying objects as cats. However, the technology is quickly improving and we are seeing more and more sophisticated applications such as self driving cars, text interpretation and face recognition.
Despite the attempts by clever diagrams, I do not think there is a satisfactory way of quickly summarizing how AIs work so for my first few posts I will be discussing the fundamentals of Machine Learning (as I understand them). This way, my blog will be accessible to anyone because my readers and I will have a shared foundation and vocabulary from which we can explore more interesting things such as Machine Vision, Natural Language Processing etc.
Disclaimer: I am not an expert. I am not claiming that everything I post should be taken as gospel. Instead, I encourage your feedback if I have made a mistake or even gotten something completely wrong! The whole point of this blog is so I can share what I'm learning in an accessible way and hopefully you guys can correct me, introduce me to cool resources, papers etc. so that I learn from you as well!
SO... the fundamentals posts are intended to be a self-contained introduction to the topic so you can understand later posts. However, I still encourage you to learn from other sources too as seeing things expressed in different ways can be helpful in learning something so complicated.
This summer (2020), I have been intrigued by the which face is real game to learn about face-related AIs. I have been looking into the state-of-the-art architectures and what their capabilities are. Once we've covered the fundamentals, my posts are going to be about sharing my notes and what I've learned about each architecture I've studied. For those of you who are familiar with my reading blog, these AI blogs will not be as structured. They will likely be my raw notes with about 20 minutes of polishing per post. The main objective is to make the somewhat technical papers more accessible.
I am really excited to share my learning journey with you and hope that you find it as interesting as I do. I want to reiterate that I hope this will become a two-way process and if you guys see something that doesn't make sense or could be more clear, then I want to know! I'd also be happy to take a look into any interesting AI topics you may have come across and would like to see a blog post on.
Thanks for reading,