I got into machine learning because I am also obsessed with how humans learn. I didn’t get very good grades when I was in middle school, so, since then, I’ve been reading books on learning psychology in order to learn techniques as well as honing those techniques while I’m working on problems or reading. But more practically machine learning is a way for software to adopt intuition that was only previously available to humans and that practicality is what drives me to learn more about it every day. Before, we would have to hard code all the details about a problem for a program to be able to employ a task. For example, if you wanted to program a computer to be able to detect someone’s mood based on facial expressions, you wouldn’t be able to. There are too many details about our face and so it would take too long to code up the algorithms to analyze all the intricacies that make up a facial expression.
Yet, why approach it that way when that is not how we learn? Aside from school, we learn by experiences, which we store as intuitive knowledge. If I were to ask you to describe to me in words what a curious facial expression looks like you would have a very difficult time talking about each and every ridge and crevice of the face. But, it’s very easy for you to show me how it looks like or show me a person who is curious. This way is much easier and relies on forming an intuition based on multiple examples. We don’t have to read countless textbooks on facial expressions and facial muscles in order to learn how people feel based on their facial expressions. We simply see the expression enough times to know what it means. This is the kind of learning power that software now has through machine learning. The future is about to be crazy.
There is already so much that machine learning can do for us but in the future this will become even more prevalent in our lives. Based on data about our behaviors, software becomes intuitive and is able to guess what we like and don’t like so it can give recommendations. As each one of us produces more data, the number of types of recommendations and the quality of recommendations is going to skyrocket. Imagine a world in which your refrigerator suggests a food based on your mood as determined by your facial expressions detected through a camera in you house, or even recommends food based on information on your current workout routine. Imagine your fridge being programmed to restrict certain foods based on learned health data and what it knows about your current health state. More controversially, imagine a world in which there is no financial crash because all the world’s financial institutions can predict them with ease.
How do you get into this exciting field? Well, Machine learning sort of works like a black box in which you put a piece of data into the black box like an image of a face, and the black box outputs a label for that piece of data such as “happy”. For this reason, in machine learning jargon that black box is called a classifier because it classifies a piece of unknown data sort of like a person sorting a bunch of unlabeled face into male or female. As you learn more you can determine the best type of “black box” to use based on the problem or go even deeper and develop your own new algorithm. So, as someone passionate about machine learning, you can either work on creating new machine learning algorithms or you can use algorithms that already exist in order to solve real world problems. Unlike in other fields though, the line between these two is blurred but still generally separated by a graduate degree. In my next post I describe the specific skills to learn it and how you can get involved today.