Browse Categories
Choose your Categories to read

What is Machine Learning

aniket verma
Nov 9, 2021

Introduction

This blog will discover what machine learning is, why it’s so popular, its importance, its applications, etc. Have you ever thought about how shopping websites like Amazon make recommendations of the products you should buy, along with other products that you are buying? 

 

Source: Link

How can Alexa, Siri, Google assistant, etc. understand what one is saying and act accordingly? These mind-blowing features and overwhelming technologies have all been made possible by machine learning and rapid advancements in the field of Artificial Intelligence.  

Let’s get started with what is machine learning?

What is Machine Learning?

The term Machine Learning was first coined by Arthur Samuel in 1959. He defined machine learning as the field that gives computers the ability to learn without being explicitly performed. It is the subset of Artificial Intelligence. It gives the machines the ability to learn more human-like behaviour which keeps on learning and developing themselves with no explicit programming. The only way humans intervene throughout the process of learning is by applying their experience of machines to get better results.

 

One might think that why is it so over-hyped and how is it different from traditional programming? The main difference is that, in the traditional way of programming we have some input data, we write some well tested code or a set of rules and according to the rules we get the output. But in machine learning we are given some data and we also have the output/results that are being fed. The task is to find out the set of rules using which we can predict the output. Thus it develops over time. 

Let’s understand it pictorially.

 

Let’s go deeper and understand some terminologies that are commonly used while we discuss anything in ML.

Some Important Terminologies

Some of the standard and important terminologies used in Machine Learning are as follows:

  1. Model: a model is a mathematical representation of a real world process that learns according to the data fed. The learning algorithm and dataset together comprise and build a machine learning model.
  2. Feature: It is a measurable property that can predict the output of a given dataset. It may be possible that the output doesn’t depend on a particular feature, but all features initially equally contribute to the output. We can later on reject or select features depending on analysis and using several different techniques like PCA etc, that will be discussed later.
  3.  Training a model: This term is often used in machine learning. Training a model refers to feeding data to a learning algorithm and allowing it to run over the fed data to learn or find the patterns over time to predict expected results.
  4. Target/Label: It refers to the output value that the algorithm/model must predict.
  5. Overfitting: What a large dataset is fed to a training model, it tries to mug up all the values rather than trying to find a general trend/pattern. This happens because datasets fed to the training model usually contain some noise. The model learns the dataset instead of a general trend which the given dataset is expected to follow, but the noise and sometimes incorrect data entries also force the model to make incorrect predictions.
  6. Underfitting:  It is opposite to what underfitting means. When a model is underfitting, it means that it cannot learn the general trend/pattern of the given dataset and cannot fit well.
  7. Outlier: It is a very common term that most students cannot understand and use blindly. An outlier is a point in a given dataset if it doesn’t follow the distribution that other dataset points follow.

Source: Link

The above image shows an example in which a general dataset points are plotted in a graph. The model(Linear regression model) predicts the general trend followed by the dataset points except the marked dataset point which does not follow the pattern as predicted, hence it’s considered as an outlier.

Some prerequisites before practicing machine learning 

Before jumping into machine learning it’s important to have proper knowledge of some common prerequisites that are as follows:

  1. Mathematical tools:
    1. Linear Algebra: Machine learning uses a lot of concepts from linear algebra. Hence, it’s expected to have a proper knowledge of vectors, matrices, norms, vector spaces, etc. 
    2. Probabilistic Models and processes: You will encounter a lot of algorithms which are implemented using a plethora of probabilistic models and processes, and are also used for analysing several results. Hence it becomes imperative to have a strong hold over probabilistic models and processes.  
    3. Mathematical analysis of derivatives and gradients: One must be aware of gradients and derivatives and know how to compute and analyse them. One should know how a particular result used in a machine learning model is used to predict the results.
    4. Multivariate Calculus: To understand the points above we must learn multivariate calculus.
    5. Algorithms for Convex optimization(some basic knowledge also works): This point is important because we will encounter convex functions and curves in machine learning and we would need to optimise them.
  2. Some programming knowledge: One must be aware of how a machine learning algorithm works and has been implemented and for that one should have prior programming knowledge and experience.
  3. Knowledge of programming Language: It is mandatory to learn at least one programming language. Generally in most of the cases people use Python, but we can also use other languages like R, etc.


Tip of Advice: Please don’t jump into machine learning without following the above points, otherwise you would not be able to understand the concepts in machine learning.  

How does Machine Learning work

The working of machine learning can be explained in a 7-step process:

  1. Gathering a dataset: The first step is to gather a dataset which is a good quality dataset. This step is the most challenging task because finding a good quality dataset is not easy.
  2. Prepare and Process the dataset: Once data is gathered, we will prepare and process the dataset. This step is needed because datasets are not readily available in the format our model’s learning algorithm will require to train upon. Even if the format is as per requirement, the dataset might have some missing values or some noise that needs to be taken care of.
  3. Choosing a model: In this step we choose a model’s learning algorithm that might give good results.
  4. Training of the model: Feed the processed dataset to the training model.
  5. Evaluation and Analysis: This step is very important because if you do not analyse the model’s performance accurately and accurately, it might happen that the model giving great results can turn out to be disastrous.
  6. Hyperparameter tuning: This step can help you improve the performance of the model by changing the parameters that are under human control for example: you can train the model for a large number of iterations so that model can give you better results. But it can backfire too, hence we have to tune the hyperparameter( learning rate here).
  7. Prediction: We finally have to predict the results. 


If we get good results and have analysed the model’s performance, we can deploy the model, otherwise we need to revise our model.

We can understand how machine learning works through the following diagram:

Different types of Machine Learning

There are primarily 3 types of machine learning:

  1. Supervised Learning: As the word suggests, supervised learning is a sub-type of machine learning in which training is supervised. In simpler words the data fed to the training model is a labelled data-set or whose output is known. The model learns the mapping function which maps the input to a predicted output and this is how supervised learning works. Problems discussed under this sub-type of machine learning are regression, classification problems, etc.  
  2. Unsupervised Learning: As the word suggests, unsupervised learning is a sub-type of machine learning in which training is unsupervised. In simpler words the data fed to the training model is not labelled data-set or whose output is unknown. The model learns on its own and deciphers the patterns of the dataset using probabilistic models to predict the outputs. Problems discussed under this sub-type of machine learning are Association, clustering problems, etc.
  3. Reinforcement Learning: This sub-type of machine learning is different from supervised learning because in reinforcement learning we make decisions by focusing on maximising the rewards or feedback of the actions that can be performed. So here also we have no specific output. So the machine learns from its own experiences. 


If you want detailed information about all the three major subtypes of machine learning you can visit the blog on different types of machine learning.

Applications of machine learning

With the advent of Machine Learning, the technology and the industries have flourished multifolds in various fields. Hence, one can find a lot of machine learning applications all around the globe. Some of the very common applications of machine learning are:

  1. Automatic Speech Recognition: You must have seen this feature in zoom meeting recordings that your speech is converted into digital text which is a great application of machine learning.
  2. Image/Face recognition: Another application in machine learning is image/face recognition. You must have seen the concept of face Id’s introduced in our smartphones. It is all possible because of machine learning.    
  3. Financial Services: An important part of machine learning is in financial services. It is used to detect any frauds or any money laundering taking place which is a critical issue. 
  4. Health Care: The technology in Health care has increased multifolds and machine learning has given a significant contribution in improving the health care systems.
  5. Recommendation systems: This a widespread application of machine learning which is adopted by a large number of tech-companies around the globe. They use recommendation systems as a business strategy.


These applications also show the importance of machine learning in our lives.

Different Machine Learning Algorithms

There are many machine learning algorithms but some of the very popular ones are as follows:

  1. Linear Regression
  2. Logistic Regression
  3. Naive Bayes
  4. Decision Tree
  5. Random Forests
  6. KNN
  7. SVM
  8. SVD
  9. PCA
  10. K-Means Clustering 


It can also be understood using the following flow chart.

Source: Link

Frequently Asked Questions

  1. What are the major uses of Machine Learning?
    Machine Learning has several uses but some of the common uses are facial recognition, health care, recommendation systems, email spam filtering, and many more. 
     
  2. What is machine learning?
    The term Machine Learning was first coined by Arthur Samuel in 1959. He defined machine learning as the field that gives computers the ability to learn without being explicitly performed. It is the subset of Artificial Intelligence. It gives the machines the ability to learn more human-like behaviour which keeps on learning and developing themselves with no explicit programming.
     
  3. What is semi-supervised Learning?
    Semi-supervised learning is also a secondary sub-type of machine learning in which the data fed to the training model is such that a large proportion of data is unlabelled and only a small proportion is labelled. Then the labels for unlabelled data are predicted using the supervised learning on the labelled data.

Key Takeaways

This article taught us what is machine learning. We discussed some important ML terminologies, its prerequisites and we saw how machine learning works. Then we looked at subtypes of machine learning and its applications.  

We hope you were able to take away a brief idea of what is machine learning. Machine learning is a field where you will grow with experience. Therefore, we recommend you to start with machine learning once you have covered all the prerequisites and you can visit CodeStudio and check out more articles on machine learning. You can check out the course of machine learning at coding ninja! Try it out and become a ninja in machine learning.

Was this article helpful ?
2 upvotes