“Machine learning”, this term seems to be very self-explanatory, but it is a whole different world in itself. If you have anyhow come across this term, and are curious to understand what is Machine learning, and some other interesting stuff about it, do not worry, since we have got you covered.
We will understand “Machine learning” in a different, and practical way. So, if you are reading this article on what is machine learning, you would probably get a thinking approach to understanding Machine learning. So, let’s dive into learning about what is “Machine learning”.
Before that, if you are interested in some information about Python language, you can simply find the article for what is Python here.
What is Machine Learning? in Simple Words
Table of Contents
Machine Learning Definition
If you have ever used some different platforms like YouTube, Spotify, Amazon, Netflix, etc., you might have experienced that YouTube is recommending videos to you, based on your watching interests, Spotify recommends songs to you, Amazon recommends some products on the basis of the buying behavior, and Netflix can recommend movies based on your interests.
But have you ever wondered about how can these products do so? Well, this is where machine learning comes into the picture. But still, the question persists, what is Machine learning? Well, we can simply understand that Machine learning is something that enables machines to learn from the data, and make predictions, or take decisions, without being explicitly programmed.
If you try to understand some formal definition of Machine Learning, it is a branch of Artificial Intelligence, which provides machines the ability to learn from data and past experience, identify patterns, and structures, and make predictions on new data.
So, from the above examples, we can understand that Machine learning is around us already in many of the use cases. Though the term is not so new but is becoming increasingly popular now.
How Machine Learning Works (broader view)
Well, when you are now quite familiar with what is Machine learning, it might have made you curious to know how does it work? Let’s understand this with help of an example.
Consider that you have shown millions of pictures of cats and dogs to a baby, telling on each picture that this is a cat, and this is a dog. So, now when the child sees some unseen picture of a cat or dog, it can easily recognize that image as a cat or dog. But how is this possible? This is because it has learned the features of some cat or dog in this case(or we can say it has learned what a cat or dog looks like).
Something similar happens with the machines as well. We feed a lot of data to the machines, which understand the patterns, structure, and make predictions, or decisions on the new data.
Let’s understand this with an example of some E–commerce websites, like Amazon. Let’s say that you want to buy something on Amazon, so you do some search on the website or application. Now, when you search for some item, you might have seen something like – “People who bought this also bought … “. This is also done with the help of Machine learning. They analyze the buying behavior of the customers and then try to make predictions that you may be interested in buying this stuff.
For example, in the shopping behavior, most of the people who bought a mobile phone, also bought wireless earbuds, So, based on this behavior, if you are buying a mobile phone, you might see suggestions for wireless earbuds as well.
Now, how Machine Learning is done is a whole process in itself, and contains several steps, starting from framing the problem, to deploying the model. So, there are several steps involved here, but let’s not get deep into that, since here, we are just getting introduced to the term “Machine Learning”. But here, let’s discuss this quite briefly.
In an abstract way, you can understand this as if we have a lot of data, which we feed to the algorithms, and the algorithms are trained on that data, and then they are tested over the new data, and if they are not so accurate, then they have trained again, and this process continues, and then we have a successful model, to which, you input some data, and it makes some prediction.
Now, if you try to know about this in detail, you would find that there is a lot more than this going on, from the problem framing to deployment, but you can consider this as a short story about it.
Please note that we are not going into much mathematical or programming details of any of the stuff discussed in this article because then it would be too much more than just understanding what is Machine Learning, But you can feel free to explore these concepts. All the concepts are very interesting to understand and implement.
Types of Machine Learning
Now, that we are familiar with what is Machine learning, and also have a brief idea about how it is done, now it is time to dive into different types of Machine learning.
If we try to classify in a broad way, based on the type of learning that is used, we have 4 categories, which are –
- Supervised Learning.
- Unsupervised Learning.
- Semi-Supervised learning.
- Reinforcement Learning.
So now, let’s try to understand these simple terms, through some examples. Note that we won’t deep dive into the concepts, but we are going to take a broad idea here so that when you go for it next time, you can be a bit more confident about the concepts.
In Supervised Machine learning, we have labeled data for training purposes. So, we can consider this as if we are showing the image, and we are also telling that this image is of a dog, and millions of other images, telling that it is a dog, or not dog, So that later, when we show it some new image, the model should recognize it as a dog or not a dog.
Since here, the training is done on the labeled dataset, with the correct output or supervision, the learning is called supervised learning.
If we move to further classify supervised learning, we can broadly find types like classification problems and regression problems. So now, let’s try to discuss them.
In classification problems, the goal is simply to predict some categorical label or class for some given input data. We can understand here, that we are trying to classify something. For example, classifying a dog or not a dog in the image, classifying an email as spam or not spam, or predicting whether it will rain or not, and a ton of other examples can be related to classification problems.
Here, we can have the binary classification, where we are having only two classes for prediction, like a dog, or not a dog. But in many cases, we can also have the multi–class classification, where we can have multiple classes, like if we were to predict the breed of dog from the image.
So, in summary, in the classification problems, we have the output as a category or some discrete value.
If you get into deep, there are many algorithms, which can be used for classification. Here are a few algorithms listed, which are more frequently used.
- Logistic Regression
- Naive Bayes
- K Nearest Neighbors (KNN)
- Decision tree
- Random forest
- Support Vector Machine (SVM)
So, if you wish, you can get into the details of these algorithms, and try them as well.
In supervised learning, we also have regression problems, where the goal is to predict continuous value as an output. This is like when we need to predict the price of a house, the salary of some person, prediction of the value of some stock, or predicting the score of a cricket match, we make use of regression algorithms.
So, in short, we can say that whenever the output that we need to have is continuous, we are solving the regression problem.
If you get into deep, there are some algorithms, which can be used for regression problems. Here are some algorithms listed.
- Linear Regression
- Ridge Regression
- Lasso Regression
If you want, you can explore and get into details of these algorithms, and try them as well.
Now, you might be familiar with the concept of Supervised learning, where we are working on the labeled data. Unlike Supervised Learning, in unsupervised learning, we do not have labeled data. So here, in this case, we just feed the data, and the algorithm has to find some patterns, and structures in the data, and respond accordingly.
For example, if you give a lot of images of cats and dogs to the machine, and that too without labels, it can find some patterns in the images, and group similar images together, like the dog images, and the cat images. So, in this case, we are clustering, or grouping similar things with each other.
So here, in unsupervised learning, we are mostly doing the following things –
- Anomaly detection.
- Association Rule-based learning
- Dimensionality Reduction.
If you wish, you can explore these things, and get into the details, trying to implement this stuff.
But at times, we might require using the unsupervised learning approach, in order to solve some problems.
As the name says most of the things, semi-supervised learning seems to be a combination of Supervised and Unsupervised learning. Here, we have some data, which is labeled, and some data, which is not labeled. We can say that a small fraction of data is labeled, and other data is unlabeled.
This is used in cases when we have very large data, and labeling too much data is difficult, so we can just label a small fraction of data, train over it, and also find the patterns in the unlabeled data, which might not be found from the labeled data alone.
This is used in many use cases, like image classification, text classification, etc. You can explore Semi-Supervised learning if you want to explore more about it, but in short, we can understand it as a combination of Supervised and unsupervised learning.
You might have an idea about Supervised, and unsupervised learning, but reinforcement learning is quite a different thing. Here, we have an agent and an environment. The agent has to interact with the environment, where it has to make some actions, and it gets rewarded for any corrective actions, and penalized for any wrong actions. The agent then precedes the environment based on the rewards gained. The goal here is to maximize the rewards received by the Machine.
Here, reinforcement learning uses a trial and error approach, where it learns in the environment through some interaction and feedback. It is not specifically told to do, or some labeled task, but here, it is learning from exploration.
Through reinforcement learning, we can train the models, play some games, or train autonomous vehicles to drive efficiently.
If you wish, you can explore a lot of things related to reinforcement learning.
What is the need for Machine learning?
For a long, Machine Learning has been around us in many use cases. But still, there can be a question that what is the need for Machine learning. So, the thing is that with Machine learning, many complicated tasks are easily possible now, and it is helping a lot in many different areas, like spam email detection, fraud transaction detection, predictions, home automation, product recommendation, and many other areas.
So, it has and is helping humans in many different terms, and surely, we can get to see much interesting stuff ahead, with many giant companies and people working hard, to come up with new solutions to problems.
The thing is that as of now, we have too much data, or we can say that we produce too much data even on the daily basis. So, processing this data, and making predictions on that, is pretty easy with machine learning.
So, in many products and services, people and companies have started using machine learning, which makes their products and services better, and helps many people in many terms.
It is an ocean if you choose to go deep, but surely it is a great field to explore and build a career into.
In this article, we tried to understand Machine learning in a different, and practical way, not going much deeper into programming or mathematics. Machine learning is around us in many use cases, like voice assistants, spam detection, fraud detection, face recognition, self-driving cars, and many others.
Again, if we were to dive into this, there is an ocean to explore and implement. Machine learning is an excellent field if you want to explore and build a career. Not just in terms of money, but also in terms of opportunities and exploration.
You can also read about Python programming language, which is one of the most popular programming languages, and one of the languages used in Machine learning.
FAQ On Machine Learning
Q: Can you define machine learning in simple terms?
Ans: In simple terms, we can say that with Machine learning, the machines can learn from the data, and make predictions, or take decisions on the new input data. It is a subfield of Artificial Intelligence, in which, we have a goal to create machines, which can perform tasks, which normally required human intelligence to complete.
Q: What are the types of Machine learning?
Ans: Broadly classifying, we can say that machine learning can be categorized into some types, like –
1. Supervised Learning
2. Unsupervised learning
3. Semi-Supervised learning
4. Reinforcement learning
There can be other parameters for the classification of machine learning, but broadly, the above four types are the types of learning, on the basis of supervision required.
Q: What is Supervised learning?
Ans: In simple terms, supervised learning can be understood as a type of machine learning, in which the machine is trained on labeled data.
Q: What are the prerequisites for learning Machine learning?
Ans: Well, if you want to learn Machine learning, as a prerequisite, you can consider learning some statistics, Probability, Linear Algebra, Calculus, and programming language(Python recommended). Along with that, you might require to learn about some other tools related to data visualization. It would take some more time, but if you are willing to learn about Machine learning, it would be worth it.