Machine Learning Fundamentals and its scope in 2020

What is Machine Learning ?

Machine learning is one of the most interesting topics to study in the field of computer science. Machine learning is a buzzword now and there is no reason that should not happen. 

We are living in the age of Artificial Intelligence where machines are becoming more and more efficient to replace human jobs. 

Let’s try to understand what machine learning is. 

For a better understanding of what is Artificial Intelligence and its usage please check my other article on Artificial Intelligence

Let’s think about a scenario where you are trying to toss an empty used bottle to a garbage can. The first attempt you made, you threw it with little extra force. You realized that you need to throw it with less force. The second time you threw it, you realized that the angle was not proper. The third time you realized you have to decrease the height a little bit and so on.

So the above is an example of the process of learning while performing the action again and again.

According to Samuel Arthur ( A renowned Computer Scientist with expertise in Gaming and Artificial Intelligence ) said that “ Machine learning gives a computer the ability to learn without explicitly being programmed ”

In simple words, Machine Learning is the ability of a machine so that its performance improvement is directly proportional to the experience it gathers.

How the Machine Learning model is different from standard programs / applications ? 

Can a Machine think ?

To answer these questions, Let’s dig a little deeper into it.

Let’s suppose I created a simple program in Python ( or C / JAVA / C++), where every time I type any number from  0 – 9 as an input using a keyboard, the program will tell which number it is as output. This program is able to understand the digits because these characters are already predefined inside the compilers of that programming language.

---

Subscribe For Updates !

Leave this field empty if you're human:

---

But what if I change the program in such a way so that I write numbers in a piece of paper with my hands, scan and save them as images in my computer, and then give them as an input to the program? Do you think the same logic will be applied to identify the handwritten digits?

This is a perfect example of “Handwritten Digit recognition “. In this case, the program extracts the features from every image ( which are also called training data), and every time a new image is fed, it tries to compare the features of the new image ( test data ) with the existing image and whichever comes to the closest of any of the existing images it will display that digit as an output.

So how did the machine was able to identify the handwritten input digit? Every time we feed a new image it improves the learning experience of the machine and next time the machine is more intelligent. After training the machine for probably like millions of data, its efficiency will be as good as any human being.

If you compare this learning process with the previous example I gave you before where you are trying to toss a bottle to a garbage can, you may understand the two processes are very similar in terms of learning.

---

---

When you show a picture of a cow to a 1-year-old baby for the first time, they don’t know what it’s called, But over time when they see it multiple times and they hear the name from their parents, again and again, their brain is fed with the name against that picture of a cow. The next time when they see something of similar shape with four legs and tails, with horns, they immediately spell “Cow”.Sometimes they spell “Cow” when they see a dog. They are learning. 

A machine doesn’t understand what numeric value it is from 0 – 9 if we provide input as a handwritten image. But because it has already been fed million times with similar images like the way the brain of a one-year-old is fed to a “cow”, it can instantly identify the image.

Sometimes when you check out some items in amazon, you must have noticed that similar items are recommended for you in your social media feeds. It happens with me every time. How surprising is that? How did Facebook know which item I searched for in Amazon?

This is a nice example of Machine learning. The recommendation engine does it all using complex machine learning algorithms running behind.

---

---

Some other examples of Machine learning are

• Weather Forecasting
• Fraud Detection in Financial institutes
• Forecasting traffic patterns from previous traffic patterns

Classification of Machine learning 

Machine learning algorithms can be classified into four different categories.

Supervised Learning 

As the name suggests, supervised learning indicates the process of learning happens under the supervision of a teacher/supervisor.

For example, we can say when kids prepare for exams, they already have a set of questions and answers ready in their brains. These were supervised by the teachers on different subjects. So when those questions or similar questions appear in the exams their brains already know the right approach to solve those problems. 

In the case of supervised learning, we already train the machine with a set of correct answers. These answers are labeled with the correct names. Now when we feed the machine with new data, it tries to figure out the new data based on the existing data with correct labels.

Let’s say,  I have already entered some pictures of different animals on my computer e.g. cow, dog, cats, etc. and I saved them with proper names/labels. So they already have their correct labels. Now if I give a new input of another cow/cat/dog, our program will try to compare what are the features getting closest to any of the existing images.

There are two subcategories of  supervised learning

1. Classification
2. Regression 

Unsupervised Learning

This is the type of algorithm where the machine is not provided with any target with labels like supervised learning.

E.g. Let’s say you went to watch a live soccer game in your life for the first time. Before that you never watched or saw soccer in your entire life. You went there because your friends forced you to. 

When the game started you had no clue initially what was happening. But over the next 15-20 mins, you noticed the following things

---

---

• There are 2 types of players. One is kicking and running behind the ball and trying to score, another one is standing at the goal post and trying not to let the ball in.

• There are two different teams with different jersey colors and playing against each other.

• Your friends are supporting one of the two teams and cheering when they are playing better or scoring goals.

• There is a third category of individuals who are controlling the game if someone is breaking any rule e,g, referee.

---

Subscribe For Updates !

Leave this field empty if you're human:

---

This way you understand the basics of soccer without anybody guiding you or supervising you and your understanding of the game will improve when watching this game more number of times. This is a classic example of unsupervised learning. 

Unsupervised learning helps to understand the patterns which were not known previously and thereby helping to categorize or finding the association or dependency mapping. This type of learning works on unlabeled data. 

There are two subcategories in unsupervised learning

• Clustering
• Association

Reinforcement Learning

Reinforcement learning is about taking suitable action to maximize rewards in particular situations. 

For example, let’s imagine there is a monkey sitting on a tree, trying to fetch a hand of a banana hanging in another tree. The monkey doesn’t know or can’t calculate initially how to reach the banana or how long he has to jump to fetch the banana. So it has to try for multiple attempts. The first time he couldn’t make the right jump and ended up landing in a hole where he got stuck and struggled to come out.

But this monkey will improve his performance over a certain number of jumps. He doesn’t know initially what is the correct route to fetch the banana.

---

Subscribe For Updates !

Leave this field empty if you're human:

---

Reinforcement learning emphasizes on mainly two aspects

• Agent 
• Environment

The environment is the setting where the agent performs an action, and the agent is the reinforcement algorithm. In the above example, the monkey is an agent, the banana is the reward and the forest and trees are the environments. 

The difference between Reinforcement and Supervised learning is that in the case of supervised learning we feed the machine with correct values/labels. But in the case of reinforcement learning the correct values are not known beforehand. The agent improves its performance based on its experience and rewards.

The environment will reveal itself to the agent in the form of states ( S ). The agent then has some influence on the environment by taking actions ( A ) and also the agent receives back some kind of reward for the most recent State (S ) – Action ( A ) combination.

Reinforcement learning has become very popular in recent days and it has a wide area of usages.

• Traffic control 
• Resource Management in Computer Clusters
• Robotics 
• Deep learning
• Chemistry 

Semi supervised learning

So in today’s date, machine learning algorithms can be classified into mainly three categories. Supervised learning, Unsupervised learning & Reinforcement Learning. 

Keeping reinforcement apart, the main focus goes on Supervised and unsupervised learning.

There are some disadvantages to both of them. In supervised learning, we have to manually label data with correct values and those data must be fed to the computer.

This is a very expensive process and time consuming.

On the other hand in the case of unsupervised learning, the chances are limited. We can say that the scope to identify is limited and depends on the learning experience. 

So a new machine learning technique has been introduced. This is called semi-supervised learning, which falls in between Supervised learning & Unsupervised learning.

---

[wp_quiz id=”1721″]

---

In this type of learning, the machine is provided with a little amount of labeled data ( training data set) and then it is exposed to a large amount of test data. Thereby eliminating the cost of labeling data manually. 

Some real time applications of semi-supervised learning are

1. Speech analysis
2. Protein sequence classification

---

Giving is not just about making donations, it’s about making a difference !! Content Shark needs your little help to keep this community growing !!