“Data is abundant and cheap but knowledge is scarce and expensive.”
In last few years, the sources of data capturing have evolved overwhelmingly. No longer companies limit themselves to surveys, questionnaire and other traditional forms of data collection. Smartphones, online browsing activity, drones, cameras are the modern form of data collection devices. And, believe me, that data is enormous.
There is no way a human can look at such huge amounts of data and make sense out of it. Even if it is possible, it would be prone to irresistible errors. Is there a way out? Yes, Machine Learning has enabled humans to make intelligent real life decision by making relatively less errors.
Have a look at the exciting ~ 4mins video below. It gives an idea of how machine learning is making computers, and many of the things like maps, search, recommending videos, translations, etc. better.
At the end of this article, you will be familiar with the basic concepts of machine learning, types of machine learning, its applications, and a lot more. Let us begin by addressing the elephant in the room.
What is Machine Learning (ML)?
The search engines (Google, Bing, Duckduckgo) have become the new knowledge discovery platforms. They have answers (probably accurate) to almost every silly question you can think of? But, how did it become so intelligent? Think about it!
In the meanwhile, let us first look at a few definitions of machine learning. The term “machine learning” was coined by Arthur Samuel in 1959. According to him,
+ "Machine learning is the subfield of computer science that gives computers the ability to learn without being explicitly programmed."
Tom M. Mitchell provided a more formal definition, which says,
+ "A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E."
In simple words, machine learning is a set of techniques used to program computers and make decisions automatically. How does it make decisions? It makes decisions by detecting (or learning) pattern in the past data and generalising it on the future data. There can be different forms of decisions such as predictions of the house prices or the weather or customer behavior, or classifications, like whether a spoken word in a recording is "world" or whether a photograph contains a face. To enhance the process of detecting these patterns and improving decision-making, one can make use of data simulation.
An ideal example for practical use of machine learning is email spam filters. Services like google, yahoo, hotmail etc uses machine learning to detect if an email is spam or not. Furthermore, there are numerous other applications that as well which we'll look at later on in this article.
+ “True loneliness is when you don’t even receive spam emails.”
What are the different types of ML algorithms?
There are several types of ML algorithms and techniques that you can easily get lost. Therefore, for better understanding, they have been divided into 3 major categories. Following is a list of different categories and types of machine learning algorithms:
1. Supervised Learning
It is one of the most commonly used types of machine learning algorithms. In these types of ML algorithms, we have input and output variables and the algorithm generates a function that predicts the output based on given input variables. It is called 'supervised' because the algorithm learns in a supervised (given target variable) fashion. This learning process iterates over the training data until the model achieves an acceptable level. Supervised learning problems can be further divided into two parts:
- Regression: A supervised problem is said to be regression problem when the output variable is a continuous value such as “weight”, “height” or “dollars.”
- Classification: It is said to be a classification problem when the output variable is a discrete (or category) such as “male” and “female” or “disease” and “no disease.”
A real-life application of supervised machine learning is the recommendation system used by Amazon, Google, Facebook, Netflix, Youtube, etc. Another example of supervised machine learning is fraud detection. Let's say, a sample of the records is collected, and it is manually classified as “fraudulent or non-fraudulent”. These manually classified records are then used to train a supervised machine learning algorithm, and it can be further used to predict frauds in the future. Some examples for supervised algorithms include Linear Regression, Decision Trees, Random Forest, k nearest neighbours, SVM, Gradient Boosting Machines (GBM), Neural Network etc.
2. Unsupervised Learning
In unsupervised machine learning algorithms, we only have input data and there is no corresponding output variable. The aim of these type of algorithms is to model the underlying structure or distribution in the dataset so that we can learn more about the data. It is called so because unlike supervised learning, there is no teacher and there are no correct answers. Algorithms are left to their own devices to discover and present the structure in the data. Similar to supervised learning problems, unsupervised learning problems can also be divided into two groups, namely Cluster analysis and Association.
- Cluster analysis: A cluster analysis problem is where we want to discover the built-in groupings in the data.
- Association: An association rule learning problem is where we want to discover the existence of interesting relationships between variables in the dataset.
In marketing, unsupervised machine learning algorithms can be used to segment customers according to their similarities which in return is helpful in doing targeted marketing. Some examples for unsupervised learning algorithms would be k-means clustering, hierarchical clustering, PCA, Apriori algorithm, etc.
3. Reinforcement Learning
In reinforcement learning algorithm, the machine is trained to act given an observation or make specific decisions. It is learning by interacting with an environment. The machine learns from the repercussions of its actions rather than from being explicitly taught. It is essentially trial-and-error learning where the machine selects its actions on the basis of its past experiences and new choices. In this, machine learns from these actions and tries to capture the best possible knowledge to make accurate decisions. An example of reinforcement learning algorithm is Markov Decision Process.
In a nutshell, there are three different ways in which a machine can learn. Imagine yourself to be a machine. Suppose in an exam you are provided with an answer sheet where you can see the answers after your calculations. Now, if the answer is correct you will do the same calculations for that particular type of question. This is when it is said that you have learned through supervised learning.
Imagine the situation where you are not provided with the answer sheet and you have to learn on your own whether the answer is correct or not. You may end up giving wrong answers to most questions in the beginning but, eventually, you will learn how to answer correctly. This will be called unsupervised learning
Consider the third case where a teacher is standing next to you in the exam hall and looking at your answers as you write. Whenever you write a correct answer, she says “good” and whenever you write a wrong answer, she says “very bad,” and based on the remarks she gives, you try to improve (i.e., score the maximum possible in the exam). This is called reinforcement learning.
Where are some real life applications of machine learning?
There are numerous applications of machine learning. Here is a list of a few of them:
- Weather forecast: ML is applied to software that forecasts weather so that the quality can be improved.
- Malware stop/Anti-virus: With an increasing number of malicious files every day, it is getting impossible for humans and many security solutions to keep up, and hence, machine learning and deep learning are important. ML helps in training anti-virus software so that they can predict better.
-
Anti-spam: We have already discussed this use case of ML. ML algorithms help spam filtration algorithms to better differentiate spam emails from anti-spam mails.
- Google Search: Google search resulting in amazing results is another application of ML which we have already talked about.
-
Game playing: There can be two ways in which ML can be implemented in games, i.e., during the design phase and during runtime.
- Designing phase: In this phase, the learning is applied before the game is rolled out. One example could be LiveMove/LiveAI products from AiLive, which are the ML tools that recognize motion or controller inputs and convert them to gameplay actions.
- Runtime: In this phase, learning is applied during runtime and fitted to a particular player or game session. Forza Motorsports is one such example where an artificial driver can be trained on the basis of one's own style.
-
Face detection/Face recognition: ML can be used in mobile cameras, laptops, etc. for face detection and recognition. For instance, cameras snap a photo automatically whenever someone smiles much more accurately now because of advancements in machine learning algorithms.
-
Speech recognition: Speech recognition systems have improved significantly because of machine learning. For example, look at Google now.
- Genetics: Clustering algorithms in machine learning can be used to find genes that are associated with a particular disease. For instance, Medecision, a health management company, used a machine learning platform to gain a better understanding of diabetic patients who are at risk.
There are numerous other applications such as image classification, smart cars, increase cyber security and many more.
How can you start with machine learning?
There are several free open courses available online where you can start learning at your own pace:
-
Coursera courses
- Machine Learning created by Stanford University and taught by Andrew Ng: This course provides an introduction to machine learning, data mining, and statistical pattern recognition. Click here
- Practical Machine Learning created by Johns Hopkins University and taught by Jeff Leek, Roger D. Peng, and Brian Caffo: This course covers the basic components of applying and building prediction functions with an emphasis on practical applications.
-
Udacity Courses
- It is a graduate-level course that covers the area of Artificial Intelligence concerned with programs that modify and improve the performance through experiences. Click here
- Introduction to machine learning taught by Katie Malone and Sebastian Thrun: Click here
-
edX courses
- Principles of Machine Learning taught by Dr. Steve Elston and Cynthia Rudin: Click here
- Machine Learning taught by Professor John W. Paisley: Click here
You can also check out the detailed list of free courses on machine learning and artificial intelligence. To conclude, machine learning is not rocket science (though it is used in rocket science). This article is meant for people who have probably heard about machine learning but don’t know what it is. This post just gives a basic understanding for a beginner. For more detailed articles, you can go here.
