An Introductory Insight Into Machine Learning

An Introductory Insight Into Machine Learning

There has been a lot of hype about Machine Learning in the last few decades and people from various fields are converging towards Machine Learning. It is undeniably one of the hottest topics today and will continue to stand on the top for at least a few decades to come. We will try to understand some concepts of Machine Learning. By the end of this article, you will have an idea of the rudiments of Machine Learning. 

What is Machine Learning?

Machine Learning, as the name suggests, is the process by which a machine learns on its own without human intervention. As a machine learns on its own, it is a subset of Artificial Intelligence. Machine Learning algorithms are used to build a “mathematical model” which is then used by the machine to make predictions or decisions without explicitly programming. As a mathematical model is built, Machine Learning is closely related to mathematics. 

This is the video version of this article

History of Machine Learning

The term Machine Learning was coined in 1959 by Arthur Samuel, an American pioneer in the field of Computer Gaming and Artificial Intelligence. Machine Learning at that time was mostly used to deal with pattern recognition. 

If you want to learn more about the history of machine learning, here is an interactive roadmap by Google of how ML came into existence.

Classification

Early classifications for machine learning approaches sometimes divided them into three broad categories, depending on the nature of the “signal” or “feedback” available to the learning system. These were:

  • Supervised learning: Where both input data and desired output data is fed to the machine to learn the pattern. As the machine learns the pattern, it creates a hypothesis function, h(x) which is a function with input to output mapping. The function is given the name hypothesis for historical reasons but let us concentrate on the type of learning. h(x) = f: input -> output. Once the function is generated, we can use different inputs to predict the respective outputs.
supervised learning
Supervised Learning (Source)
  • Unsupervised learning: Where we don’t know the outputs and we allow the machine to find a structure/pattern by feeding just the input. It is used to find the hidden patterns or cluster the data into segments.
unsupervised learning
Unsupervised Learning (Source)
  • Reinforcement learning: It is used in a dynamic environment where a machine does some work and is either rewarded or punished based on its move. Punishment here does not refer to physical penalization. It is usually feedback in the form of a score. As the feedback is given, the machine tries to learn from it and maximize the rewards. It is mostly used for playing games, driving a car, etc.
reinforcement learning
Reinforcement Learning (Source)

Areas where Machine Learning is used

As I have already stated that people from different fields are learning Machine Learning, we can estimate the versatility and the wide range of fields it is being implemented in. 

  • Virtual Personal Assistants: Most of us are familiar with Siri, Google, Alexa and have experienced their usage by saying either “Hey Siri” or “Ok Google”… Nowadays, AI chatboxes also have an increasing demand. They are being used in many business websites to answer general FAQs, welcome, and other small talks. These assistants are used by people of different ages from children asking for homework problems, to parents listening to the latest news to grandparents listening to music. You might wonder what the machine is learning here. Your speech is converted to text and broken into tokens(the smallest elements) and then it performs the actions based on tokens. It is a process of continuous learning where the algorithm takes the input data(speech) and learns your pronunciation and modifies the tokens for better results.
  • Predictions while Communicating: When we navigate using Google Maps or a suitable GPS app, we get the directions, traffic, and also the estimated time. When we book a cab, we get the estimated price according to the demand and also the waiting time. Machine Learning plays a major role here.
  • Social Media Services: We get our personal feed and also recommendations of people me might know and also an option for naming a person in a photo and many more. Machine Learning uses the simple concept: understanding with experiences. Facebook continuously notices the friends that you connect with, the profiles that you visit very often, your interests, workplace, or a group that you share with someone, etc. And thereby suggests new friends, posts. The face recognition algorithm is used here to recognize faces and suggests names from your friend list. 
  • Video Surveillance: A single person monitoring all the surveillance cameras at a time is impossible. This is where Machine Learning comes into the picture. With computer vision and other techniques, the machine can look for malicious activities and warn the authorities beforehand. It can calculate the distance between people to ensure social distance is maintained during this pandemic. It is used to detect any suspicious activity.
  • Email classification: You get hundreds of emails daily and they automatically classified into the respective category(spam, important, junk). How is it being done? Who is doing that internally? Machine Learning is the answer. It uses supervised learning (input to output mapping) to learn what is spam and what isn’t. The same model can be used to detect malware.
  • Product Recommendation: You do a lot of online shopping and Amazon, Flipkart suggests you some items. How is Amazon/Flipkart able to know our recommendations/choices. When you add an item to a cart or mark it for later use, the machine learns your choice and recommends based on your previous purchase for a better experience and marketing.
  • Ad Marketing: When you browse the web, you find many advertisements particular to you. The machine learns from your past searches and clicks and shows them after refining. Machine Learning is used by companies for better profits on advertising by using Upper-confidence bound or Thompson sampling algorithms (reinforcement learning).

There are many other applications like YouTube recommendations, Google search results optimization, self-driving cars, Robotics and many more to come. Maybe one day, when you inspire from this article and start learning Machine Learning, you may develop an application. 

How is Machine Learning implemented?

Till now we learned what Machine Learning is, its history, its applications. Now let us discuss how they are implemented. 

To implement a Machine Learning model, we need data. More the data, the more accurate predictions we can make. Once we get the available data, we clean the data to be used for training the model. Then we divide it into the training set on which the algorithm is implemented and the test set on which the predictions are tested. Usually, a 0.75 ratio is maintained. i.e., ¾ parts for training data and ¼ parts for testing the model (However, we do not split the data for unsupervised learning). This is called Data Preprocessing.

Now the data is trained using a particular algorithm for a particular problem. Once the data is trained, we can now predict new values or test our model’s performance on the test set. We will see how to predict the performance in later articles. 

While training the data, two extreme conditions have to be taken care:

  • Underfitting: Where the model is unable to fit the training data well. It results in poor performance on the test set. The model is also said to have “high bias”.
  • Overfitting: When the model is so accurate on the training set that every data point is perfectly mapped. Even though it works extraordinarily on the training set, its performance on the test set is relatively poor. The model is said to have “high variance”. 

There are many ways to avoid overfitting (regularisation where the value of the parameters is reduced to a great extent). 

Once the model is trained and tested, it is ready to be deployed. 

What is the future of Machine Learning?

Let us see, what kind of advancements we can expect in Machine learning in the future and how would it help us.

The future of Machine Learning is unpredictable but the future is Machine Learning. Click To Tweet
  • We are able to use ML in medicine to find if a tumor is malignant or benign, find any breakage in bones and issues related to heart, decode the structure of a nucleotide, and many more. With the introduction of robotics, many typical and complex surgeries are performed at a 99% success rate with robots. There will be an advancement in the field of ML in Medicine where people can find out the disease by themselves, get proper suggestions, get internal changes report, and many more so that patients need not visit a hospital every time. It basically becomes a family doctor. 
  • Driving has become a hectic work for many. There are more road accident deaths per year than deaths due to some pandemic. Either people are not experienced or are reckless. Both are equally dangerous. So Elon Musk, a visionary developed self-driving cars. But way before that, in 1939, Norman Bel Geddes created the first self-driving car, which was an electric vehicle guided by radio-controlled electromagnetic fields generated with magnetized metal spikes embedded in the roadway. That concept kept on evolving and now, Tesla is able to produce self-driving cars. In the future, almost every car would be a self-driving car with more capabilities like being able to go on the water. You might think this as fictional. But they are being developed in reality. Even self-driving drones will be introduced where the drone delivers the package to the customer by air without getting delayed. 
  • Robots, hm. Who doesn’t like robots? Who doesn’t like their work to be done? Robots have evolved over the decades. They are now capable of replicating human actions like cleaning, cooking, playing games, gaining knowledge, performing surgeries, and so on. In the future, robots will take part in wars(if any), find out the medicine to a disease on independent research, will be sent for space expeditions, and many more tasks.
  • In the future, we need not read a whole book to note the important points. When we pass a book as an input to a model, it goes through all the words and does it for us. If there is a video, it can even give the highlights.

There is a lot of progress going on in ML. Everything will be automatic and easy for us within the next few decades. 

There is a hype that Artificial Intelligence will take over humans. That is not true. We will discuss it in the upcoming articles. 

For now, I hope you got a brief idea of what Machine Learning is and how it affects us. Are you interested? Then start a career in Machine learning for your field of study.

Recommended2 recommendationsPublished in Blogs, Technology

1 thought on “An Introductory Insight Into Machine Learning”

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top