What exactly is Machine learning?

Your quick and easy cheat sheet

Every day new reports are coming out referencing more companies investing in Machine Learning, not to mention new applications and uses. Between all of that, what ML is sometimes gets lost in the shuffle. It’s not a wizard or anything otherworldly, but it is squarely in that “magic” realm between the interface we see and all that goes on behind it.

My hope with this post is to teach you some of the magicians tricks in order to understand the capabilities and opportunities for the technology.

First off, Machine Learning does not equal A.I. They’re similar, and I can certainly make a whole post about their differences. But for the sake of brevity I’ll explain it like this. Machine Learning is about gaining insight or knowledge, A.I. is about acquiring wisdom. ML is focused on making the most accurate solution, where A.I. is focused on how successful it is working on it’s own.

Summarizing it in an analogy, A.I. is a self driving car, taking in multiple inputs and working on its own. Machine Learning is Google Maps suggesting a faster way home that you never thought of.

But, what is it and why should I care?

Machine Learning is nothing without what powers it; data. You can build a robust application, but just like a car with no gas, it won’t go anywhere without data.

Now there’s many different types of machine learning that can be leveraged many different ways, with many different algorithms or frameworks. I’m intentionally speaking theory as to not get too tripped up on details.

That being said, let’s build a theoretical traffic app to help a delivery service work more efficiently. Our local city was kind enough to give us a real time stream of all their traffic data; length of lights by intersection and direction of travel, traffic patters, accident reports and weather data.

With that, we can build an app leveraging machine learning to find insights, or what I like to refer to as those little, ‘huh, that’s interesting,’ moments. How we go about doing that is to first split our data into two categories, a “training” category and a “testing” category.

Training is about what it sounds like, it’s where our ML app will ingest the data and try to predict an outcome based off of it. For our app, let’s say it’s trying to predict traffic patters for time of day and weather conditions. We choose an algorithm, set the program off to train from 70% of the data and from what it’s ‘learned’ predict what the outcome would be given ‘x’ condition (in our case, time of day and weather condition).

Testing, again fairly straight forward. We take the other 30% of the data as testing data and see how accurate our program is at ‘guessing’ the right conditions. All software development is iterating, but ML especially is. So from here, you train and test, again and again, trying to get as close to ‘perfect’ as possible.

Once your result is near enough to 100% as your comfortable with, the real fun can begin! As I mentioned before, ML is all about insights. In the case of our traffic app, it can come back and predict that there’s a high likelihood given a weather report for an accident in a the intersection of Grand and Milwaukee. Or if you’re looking to change traffic signals, what the likely effect would be on traffic.

Machine Learning gives you a great model to test changes, predict likely outcomes or gain insight into how efficient processes are running.

What if it wasn’t traffic

What if it instead was patient data and you were trying to predict the best course of treatment? The probability that you’d win a lawsuit? Or what customer segment would be most likely to purchase your new product?

Machine learning has already entered our life in obvious ways we don’t notice. It’s started with search, guidance and advertising. The next step will be for it to intentionally and specifically tackle a complex enterprise, industry or use case. Whether it’s for the companies or consumers, it’s benefits are numerous and I hope from this you’ll be able to see where else the technology can go.

Machine learning = insight and efficiency

A.I. = Wisdom, and for that I’ll cover some other time.