Header Ads

Different Types of Machine Learning Explained

Different Types of Machine Learning Explained

I'm gonna be doing is sharing with you my top 4 machine learning techniques that you need to know. Now there's so many different ways of doing machine learning and artificial intelligence but these are kind of the 4 main methods and you need to understand what these methods are so you can appropriately pick the right one for whatever project you're working on. It's one thing to go through tutorials and to watch other people apply specific algorithms to specific projects.

But to actually be able to pick which algorithm that you're going to use for your own project is its own skill. And hopefully, this article will help you do that.

So I want to make a clear distinction just to start the article here on what I mean by machine learning. A lot of people confuse artificial intelligence and machine learning. And to be honest with you the Internet does not make this easy to kind of distinguish between. A lot of people will label their stuff as machine learning when really it's a different artificial intelligence strain or they'll say AI and it's people get really confused. So Let's clear it up. Right now.

Machine learning is a strain of artificial intelligence. Artificial intelligence is kind of the top tier and everything kind of trickles down from there where machine learning is its own kind of topic. Within artificial intelligence, you can have other things like evolutionary algorithms or you could have different kinds of AI algorithms that do specific things that are in their own little bubble.

Machine learning is one of its own bubbles that strains from artificial intelligence. Now the topics I'm going to be talking about here and the methods are kind of strains off of machine learning. So the different types of machine learning they're not necessarily algorithms. There's algorithms that are involved with these methods but they're kind of their own strain and different algorithms will fit into that. So with that being said let's talk about the first algorithm.

1. Reinforcement learning

Now reinforcement learning is a very popular method for training AIS to play video games and that's kind of the example I'm going to use to explain how this works. Now reinforcement learning is learning by trial and error. And essentially what that means is our AI starts off being extremely stupid having no idea what to do. It's seen no prior data before and doesn't know what the level looks like. It doesn't know anything and it starts taking random actions. So this is where I'm going to introduce to us three concepts an agent an environment and a reward.

Now our agent is going to be our artificial intelligence. And in this case, when we're talking about a game it would be the player of the game or whatever AI you're trying to train. Our environment is going to be if we're talking about a game may be the level at which this agent is playing. But in other situations, your environment is kind of what you're exploring and what you're trying to optimize or do the best in. And then we have a reward and the reward is what our agent is trying to maximize.

The agent is going to be navigating the environment and based on the actions it takes it's going to be given a reward. That reward will come from the programmers. They will determine what reward the agent gets at what specific scenario for doing what specific action. And essentially the goal of the agent is to maximize this reward. So it starts navigating the level kind of figuring out what it needs to do and navigating the environment.

And as it does this it starts to realize based on the actions it takes which actions maximize the reward. So if we're talking about maybe a Mario game for example Mario will start off maybe running left. So going out of the level and it will be given a negative reward and told don't do that. Stop doing that.

So the next time this happens it's gonna go the other direction because it says Well I know left is bad. That minimizes my reward. So where can I go to maximize it? So it'll start going right and then maybe it hits an obstacle. It's gonna be okay. So when I get to that point rather than hitting the obstacle which gives me a negative reward let's go around the obstacle. Let's do something that gives me a positive reward. And this is the basic premise of reinforcement learning.

2. Supervised machine learning

Now a lot of people classify this as kind of the most basic and most boring form of machine learning and I would agree but that also makes it one of the most useful. Now a lot of people that get into machine learning think that you need to use very advanced models. You need to use neural networks and crazy evolutionary things to solve problems when in reality a lot of the problems that you have can actually be solved with supervised machine learning models.

What is supervised machine learning?

Well, you start in supervised machine learning with some previous knowledge or some previous data. That is step one. You're not going to be doing supervised machine learning unless you have some data to work with. And what you're going to be doing is essentially trying to come up with some function that can map the input of your data to the output of your data.

The reason you want to do this is because you have all this previous data and you want to come up with some kind of correlations between them. So you want to say Hey maybe if a student has high grades that would relate to them studying for a long amount of hours. So what you're going to do is say okay we have all this previous data. We're going to feed it into this machine learning algorithm and it's going gonna come up with this function.

And your goal is with this function that when you pass it some input data that it's never seen before. So some data that you're trying to make a prediction for it can give you an accurate prediction for that specific value. So that's why you need a lot of data usually to make this work. Well, so what you're going to do is feed the input of this data into the model. It's gonna look at it it's going to spit out some output.

It's going to compare that output to what the output should be and then say Okay I need to make a tweak. Here a tune here to make this function a little bit better. And this is why it's called a supervised machine learning algorithm because you can almost think of it as like a teacher teaching the model what to do and how to get better where it starts not being very good.

But as it sees more and more data the teachers like No do that. You need to fix that tweak that it gets better better better better until you reach a point where you're satisfied with the accuracy of your model. And that is supervised machine learning.

3. Unsupervised machine learning

So in supervised machine learning what we had was some input data and we actually had the answer to that input data. So we had for example all this information about a student and maybe we're trying to predict their final grade. We actually had a bunch of previous student's final grades. So when we ran through the model we could tell the model when it got something wrong.

Hey, you were one grade point off from what this grade actually should be based on your current state. So Let's tune you a bit. Let's fix you a bit. Let's get better. But in unsupervised machine learning, we don't actually have the answers to our data. We have a ton of different data and we actually don't know what the answer is for what we're looking for. The problem itself is slightly undefined.

And what we're trying to actually do with unsupervised machine learning is have the computer or the algorithm come up with some correlations between our data that we can't see ourselves. So maybe we're trying to predict something with supervised machine learning. We want to figure out what grade is going to be or we want to cluster this information together.

We want to classify something and we actually know the correct answer whereas here we want to come up with things about our data that we don't know. So this is where we talk about something called a cluster. Now I think I made a mistake and I mentioned clustering with supervised machine learning that is not correct.

Clustering typically is going to be an unsupervised machine learning and a common algorithm in unsupervised machine learning is K-means clustering where maybe we feed a bunch of data points. We only have input data. We feed a ton of data points to our model or algorithm and what it does is actually clusters and figures out which data points are related together.

So maybe we're trying to figure out what you know the information will relate someone to be a part of a specific ethnic group or something like that. We're looking at some big data. Well, we don't know how many groups we might have in our data how they might be. We don't want to specify those. So we just pass this information into the computer. It does some of this magic stuff with the unsupervised machine learning algorithm and spits back to us a bunch of different clusters of this group.

And when we pass in in a new data point that we haven't seen before well what it does is say okay you are a part of this cluster or you are a part of this group. So unsupervised machine learning algorithms are typically used to try to figure out stuff about the data. We're trying to determine different groups or different kind of pieces of information and some abstract correlations that we maybe can't see ourself as a programmer or as someone that's analyzing the information.

4. Deep learning

Now deep learning is by far one of the most kind of looked up to and just like advanced parts of machine learning. Now a lot of people that get into machine learning just immediately jump into deep learning. They see all these neural network things on YouTube. They see the neural network, neural network, neural network. And they start just learning it and doing it. And this is great. And you can do this.

But deep learning is really not applicable to a lot of different tasks. Or maybe it is but it's way overkill. And you could solve a lot of the same tasks that people are doing with deep learning and neural networks which is basic supervised machine learning. Algorithms or reinforcement learning are simple. Very basic things to set up and to work with.

So what is deep learning?

Well, I've kind of said it already but it's pretty much just working within their own networks. Now the reason we call this deep is because a neural network has more than one layer so whereas in our other algorithms we kind of just had an input and an output and we had a function that kind of classified or did something with that with a neural network we have layers.

So we have an input layer, we have some hidden layers in the middle, we have an output layer and our data actually gets passed through all of these different layers to pick out specific high-level features of it.

Different Types of Machine Learning Explained

Now, this is why neural networks are used a lot for classification tasks related to things that have patterns in them. So for example, doing object detection or image recognition a classic example is going to be with the amnestied data set what essentially you're hoping a neural network is gonna do when you train it on a data set like an amnesty data set which essentially is a bunch of handwritten digits and you want to figure out if there's zeros or if they're ones 2 through 9 is.

You're gonna pass all the pixels of this image in and then the network's actually gonna look at these pixels try to pick out like lines and features and curves and edges of it and what the number is based on seeing all those different features that it's kind of created and figured out.

Different Types of Machine Learning Explained

So neural networks are really good at looking at higher-dimensional features that you can't specifically see in the data set or that actually doesn't exist it will create those features for you based on the input that you're giving it. But this means that it's very complicated it's very complex. You don't usually understand how you're actually solving the problem when you're using a neural network because you kind of just give some input you get an output and you don't have any idea what happens in between.

And that's the dangerous thing when you run into an issue where your neural networks not performing correctly. It's very difficult to tweak it and modify it and a lot of the times there's not much you can do because you really just don't understand what's happening in the middle.

So these have been kind of the 4 main topics in machine learning. Now I always like to say that I am not an expert there's a lot that I still need to learn I'm a student myself and this stuff is really interesting to me and that's why I go out of my way to make articles like this and to teach other people because I want them to be able to learn the same things that I've done.

I find a lot of people on the internet what they do is explain things super complicated and just try to make it seem like they're very smart that's really not how I want to come off. I want to explain this to you guys as simply as possible so that you can you know go out there and learn it yourself.

A lot of this stuff is really not that hard it's just you have to have a little bit of knowledge and know where to look to find the information. So with that being said, that has been the article if you guys learn something you'd like this leave a comment down below and share this article with others.

No comments

please do not enter any spam link in the comment box.

Powered by Blogger.