What even is machine learning…?
You hear these buzzwords all the time:
machine learning (ML)
artificial intelligence (AI)
data science (DS)
data analytics
But what do these words actually mean? They’re not as complex as you might think.
In my opinion, machine learning, artificial intelligence, data science, and data analytics all mean the same thing [I say “opinion” because I’m sure there are people out there that strongly, like strongly, disagree]. At the end of the day, these concepts are just describing the use of data to accomplish some job using statistics and probability to their advantage.
My favorite definition of machine learning comes from Professor Kilian Weinberger of Cornell University who states:
Machine learning models are algorithms that improve on some task with experience.
Simple as that.
Let’s define the bolded words in this definition:
An algorithm is just a recipe for decision making. The same way you follow instructions when baking a cake, a computer can be told to follow certain instructions as well. An example of an algorithm is something as simple as computing a factorial (remember those?). If I tell a computer, “hey, in order to compute 25 factorial, just start at 25 and then multiply that by 24, then multiply all that by 23, then multiply all that by 22... all the way down to 1”, that is an algorithm! I gave the computer step-by-step instructions to solve my task.
A task is just a problem we are trying to solve. In the example above, the task is creating a factorial and the algorithm is the steps to get there.
The way humans think about experience is similar and different from a computer. To a person, experience means spending time learning about the world: whether it’s a job or speaking a new language. A computer’s version of experience is data. The higher the quality or quantity of the data, the more “experience” a computer is able to learn from. Thinking about it this way, it’s actually not too different from humans! As we grow and experience life, we really are just using our five senses to collect information around us (data) and we use our brain to process the information and learn.
It is this learning process that causes us to improve and grow as humans, and machine learning models do the same. With more experience (data) they improve - they learn and get better at accomplishing the task at hand. As more or better data comes in, algorithms can improve, and tasks become more easy to accomplish.
That’s really all machine learning models are - yes, it gets much more technical with mathematical properties and complex statistical techniques; but at a higher level this is all that is happening.
For the context of this blog, I am building a machine learning model that is tasked with predicting NFL score outcomes. I built out a web scraper to collect team data, cleaned it, and engineered features for predictability. Throughout the season, I have a pipeline set up so that each week, new data can flow in and give my model more “experience” and teach it to learn and get better.
I will be posting predictions on my instagram page @bettor_picks each week using this model - plan to post 24 hours before each game time. I also plan to post comparative results between my predictions and the true outcomes on this blog each Tuesday throughout the season.
If you want to learn more about machine learning, I would highly recommend the YouTube playlist linked here. It is a machine learning course taught by Professor Kilian Weinberger at Cornell University. It is incredible that this is free on the web - he is extremely clear in the lectures and it is actually fun to follow along. Disclaimer - by lecture 3 it gets very technical - but still, a very good resource.