About Machine Learning

KALM-150x150"A brief explanation of Machine Learning, it’s applications, and how it’s developed.

Featuring Tom Merritt.



A special thanks to all our supporters–without you, none of this would be possible.

Thanks to Kevin MacLeod of Incompetech.com for the theme music.

Thanks to Garrett Weinzierl for the logo!

Thanks to our mods, Kylde, Jack_Shid, KAPT_Kipper, and scottierowland on the subreddit

Send us email to [email protected]

Episode Script
Machine Learning

AI. It has AI! If Artificial Intelligence takes our jobs, AI AI AI

When people talk about AI, a lot of the time, I’ll venture to say most of the time, they’re talking about Machine Learning.

But what is Machine Learning?
And what kinds of things are AI but are NOT machine learning.

I’m Tom Merritt and I’m going to help you Know a Little more about Machine Learning

Machine Learning is real. It’s out there and you use products that use it every day. Yes it’s used in automated cars, but also online recommendations from places like Amazon and Netflix, discovering what people are saying about your company or product on Twitter and fraud detection.

Machine Learning is not exactly an algorithm, They’re not the same thing. Machine Learning – or ML as I’ll shorten it sometimes here –is a computer system that USES algorithms and statistical models to carry out a task without explicit instructions. One of ML’s advantages over Statistics is that you can find structure without having to know what the structure should look like.

So on the one hand it is NOT generalized intelligence. It can’t just learn everything and become super smart. It has a task. It won‘t go beyond that task.

But it’s also more than an algorithm. Because instead of telling the system how to perform the task with a set of instructions, Machine Learning is just given some tools and the task and figures out how to get the right results on its own.

As defined by Tom M. Mitchell, who literally wrote the textbook Machine Learning– the system learns from experience to perform a task. If the performance of the task improves with experience, it’s machine learning.

ML starts with training data. Let’s say a bunch of pictures of cats. The ML uses its algorithms to create a mathematical model of what it thinks a cat is based on the training data. So in other words the algorithm doesn’t say “cats have whiskers, pointy ears, etc…” The algorithm’s look for commonalities and refine them until it has built a model of what a cat looks like.

There are lots of versions of Machine Learning. The two big ones are supervised and unsupervised learning.

Supervised Learning tells the ML what the right answers are. So the training data would have pictures of cats and pictures without cats and the pictures would be labeled cat or not cat so the ML knew which were cats and which weren’t. It would then look at all those pictures for patterns and create its mathematical model of what a cat is. It then applies its model to the data set and sees how well the model works at matching the labels. It then updates the model based on what it learns and does it again until it gets really good. Then you can start giving it unlabeled pictures and see how well it does at identifying which ones have a cat in it.

The second big category is unsupervised learning. It still has cats and not-cat pictures but no labels. So the ML still looks for patterns in the data and just groups them together. The idea is one of those patterns it finds will be “cats.” The ML sort of discovers what a cat is all on its own. We label that pattern it identifies as cat and tell it to keep identifying those. The ML does not really know what a cat is or that it’s called a cat. Unsupervised Learning is really good at segmenting things and identifying clusters of stuff. So customer data for instance. Since it can find all the patterns it can do things like say all these customers seem to like cats, these don’t and these are ambivalent.

As I mentioned these are just the two biggest categories of ML. There are all kinds of subcategories to these and even mixtures of the two.

Semi-supervised learning has some of its training data labeled but not all. So it mixes supervised and unsupervised strategies together. This one is used a lot in facial recognition.

Reinforced learning, used in autonomous vehicles and game playing, gives positive or negative reinforcement for correct and incorrect answers. So if, for instance, it wins the game, it knows that mathematical model worked better than the one that lost.

What approach a project uses depends on what you want the ML to achieve.

Now you see the importance of the training data.

If we train our ML on a data set that has almost entirely tabby cats, it might start to label calico cats as not a cat. So you need good training data. The better the data the better your system learns. You need good algorithms to help the system recognize patterns and evaluate data. And you need horsepower. The ML needs to go over it’s results over and over over over again. Hence the need for high powered chips at scale.


Now that we have a better idea of what Machine Learning is, let’s get back to AI. What’s an example of AI that is NOT machine learning.

Data Mining. Machine Learning and Data Mining overlap a lot and some argue they are part of the same broad discipline. Or that machine learning is part of Data Mining’s superset of tactics. But Machine Learning generally wants to reproduce known things, like which pictures are cats. We know what a cat is, ML is able to duplicate that knowledge. Data mining’s goal is to find unknown things. We didn’t know that all black cats are really good at poker. This by the way is not true – as far as I know– but it would be the kind of thing data mining might uncover for us.

Neural Networks are one way of operating machine learning. They’re not a type of ML

Deep Learning is usually described as a type of unsupervised machine learning but it’s on a whole other level. It basically takes large amounts of data all he computer power it can muster and special neural networks.

So there you go. This will not teach you everything you need to know about ML but the purpose here is when somebody mentions Machine Learning, you now have a much better UNDERSTANDING of what that means and how it’s different from together things it gets lumped in with.

In other words, you know a little more about Machine Learning