Types of Machine Learning Explained | Beginner-Friendly Guide with Real Examples

Key Takeaways

Hello everybody and welcome to an explanation of types of learning with respect to how a machine learns. So types of machine learning and the first is supervised learning Supervised learning here you could actually see it as let me explain before I write it. So supervised learning is a type of learning where we train the model with label data. So let's say for example this is our data set. So, this is an image of a dog and then an image of a cat. To help the model understand what a cat is or how a cat looks like and how a dog looks like, we label each of them. This is a dog and then we say this is a cat. So basically this is a label data. So this is in the context of if we want to train a model on how to classify images. So we label each of our data points. Data point is one subset of the entire data. So one point of the data we label it if it's a cat, cat, if it's a dog, dog, if it's a penguin, penguin and all like that we label it. That's a label data. So we feed this label data into the model. And the model learns the data by mapping the input. The input will be the image and the output will be the label dog. So it learns the relationship between the image and what makes it a dog. That's what labeled data is. So let me just write it label label data and examples of this are I'm write this in another color let's say examples examples image recognition or image classification speech recognition. Okay, so let's go to the next. We and we have loads and lots of examples. Of course, it's more than two, but I'm just restricting it to two to help you understand the context. So to revise and recap again, supervised learning is a type of learning where we train our model with labeled data. And labeled data is a data or data set that has input and output. So the input is the image of the dog in this context using image classification. The image and the output is dog. So we fit it in with the image and we teach it that once you see an animal that has this features it is a dog or it is a cat and so on and so forth. And in context of speech recognition especially for transcription. Let me even put that into especially transcription transcription or transcribing. So imagine exactly as I'm talking like this. So it's recorded like this and then somebody or a human being annotates it that is transcribes what I'm saying my spoken words into written context. So that's a label data. So the machine or the model is fed with the audio of me speaking with my spoken word that's the input and it's also fed with the answer. So see as a question and answer the answer is the written or transcribed words. So that next time when you fit it with the spoken word, it knows how to write it in written context. So see label data as question and answer. The question is the image or the spoken words as the context may be and the answer is the transcribed words or the tag or the label. So let's go to the next type of learning which is on surprised learning. So unsupervised learning on the other hand is where or involves training the model with unlabeled data. So as I put it unlabeled so you can imagine how does an unlabelled data look like. So let's just try to visualize it. So imagine this is our data set and all of these are data points. in our data set. So because they are not labeled here, our model is stacked with the responsibility of learning patterns in the data set on its own. Since it doesn't have the spies like this, doesn't have the input and output, doesn't have the question and answer, input and output, it has to learn the data on its own, try to understand it and try to extract any statistical references it can. So that's the unsupervised learning. And examples of this, a very popular example of unsupervised learning is You might have heard of it. Examples is clustering. Anomaly detection and to mention a few there are loads and loads of examples of it. So let me explain. So clustering for example we maybe we have maybe this is a data set of a store's um daily sales or weekly sales of customers or maybe for the year 2024 this is their data of all the sales they've made. So we could just input this data set into our model and tell it to cluster based on customer behavior similarities. So it just clusters based on customer similarities. And for example, it could say uh this set of customers buy on a weekly basis or these other set of customers spend this so amount or this set of customers buy this certain kind of product and so on and so forth. an anomaly detection. You could even use it in um cyber security protection where the model based on the data you give it, it's able to say this kind of behavior is an anomaly. It's not what is common within the data set. So you could use unsupervised learning for that. And basically generally unsupervised learning is for unlabelled data. has data that does not have labels or tags on them. So, let's go to the next one. And the next one is semi super vice learning. So this type of learning is actually a very tricky and interesting one because it combines because it combines both or it leverages the strengths of both supervised and unsupervised learning. It uses both labeled and unlabeled data. So let me put that here. both labeled and on labelled data. How do I mean? Okay, so let me let me try to give a visual representation of this. So this is our label data. for example, dog cat. And then then draw the cat dog. Okay. Apologies for the drawings of the cats and dogs though. Okay, so now this if this is how or the kind of data a semi-supervised learning model uses. This is labeled try and this is unlabelled. So this is labeled. But basically this is used especially in in context where label data is scarce or it's going to be really capital intensive both human wise and moneywise to obtain the label data. So the model is first trained on label data on the small or limited amount of label data then so see the label data training as the foundation and the first step of training afterwards is now trained on the larger amount of data which is then the unlabeled data. So an example of this is so the examples would help drive the point home speech rug me speech recognition. So first of is trained on labeled or annotated data. So as I'm speaking the spoken words of mine would be transcribed by human beings into written text and this would be done for hundreds or thousands of other spoken words. So now the model is given this transcribed spoken word into written text trained on it so that it understands the meaning of specific words and phrases and how they're used. And then it is given the larger amount of unlabeled speeches which is spoken words for it to decipher on its own to continue training on its own after it has gotten the proper foundation of being trained with labeled data. So it gets the foundation, the strength, the understanding of what words and phrases mean from the labeled or annotated data. And then the larger data set which is the unlabelled data in this case helps it to understand nuances and speeches, understand context, understand accents, you know, let it explore on its own and be able to get or gather or connect the dots the way a machine knows how to do best. So semi-supervised learning leverages the strengths of both labeled and unlabeled data. I hope I was able to to explain that better. Another example is medical imaging. Medical imaging still follows the explanation of speech recognition. But this case you first of all give it labeled data the limited amount of label data where the images of scans and um biopsies and all of those stuff are labeled so that the computer or the model has a foundation on how to interpret the unlabeled data. Okay. So to the last and not the list, we have reinforcement learning. So blue D we in force man learning. Okay. So this reinforcement learning is it's kind of fun to me because I see it as a gamification. So let me explain it before I go into write write it. So reinforcement learning how we train the model. We train the model to make decisions in an environment. So we simulate an environment where we train the model to make decisions by rewarding or punishing it. Let me explain. So see the environment as this and look at our AI agent here. So if it goes this way, we reward it. If it goes this way, we punish it. So let me even reward punish it or ben out it. So we train our model to make decisions by rewarding or punishing which by the radic college penalties. So if our AI agent goes this way and it's not meant to go that way, we what? We punish it by giving it a penalty. And if it goes this way and it's meant to go that way, we reward it. So that's how it is. What are the examples of reinforcement learning? First is autonomous vehicles. This is how autonomous vehicles are trained based on reinforcement learning. If it goes left when it's meant to go right, it's punished. If it doesn't stop when it's meant to stop, that's when it sees maybe a passer by and it's meant to press the brake and it doesn't, it's punished. If it does, it's rewarded so that it improves itself by hitting more towards actions that would lead it to be rewarded. Another example is game playing a jet such as chess and alpha go if you've heard of go the game go. So this is how those AI models or agents are trained. When they make the right move, they're rewarded. And when they make the wrong move, they are punished. And that's the end. Let me know if you have any questions and see you till next week. Bye-bye.