Why one should learn Machine Learning?

We are in to digital era and one buzz word is booming right now in information technology is "Machine learning and Artificial Intelligence". This blog is all about introduction to Machine learning and Artificial Intelligence. The intention behind this blog post is to make an individual aware of ideology behind Machine Learning and urgency of having it in each IT industry.

"Machine learning is the sub-field of computer science that gives computers the ability to learn without being explicitly programmed." - This is what Wikipedia says about machine learning and it is absolutely precise.

In a MORE technical term we can say like this “A computer program is said to learn from experience E with respect to some task T and some performance measure P, if its performance on T, as measured by P, improves with experience E.” -- Tom Mitchell, Carnegie Mellon University

I would like to give you one simple examples about machine learning:

Product Recommendation: When you visit any item on Amazon, Ebay and Alibaba, it learns your choice, It shows similar products right next to it and even send you mail if similar product arrives in stocks. Similarly Facebook recommends people and pages and google recommends search results.
Face Detection: A known application is, when you upload a photo on Facebook, it automatically recommends names of person present in photo.
Digit Recognition: Universal OCR technology is used for inbound and outbound international mail sorting in Germany. This application actively exploits and benefits from a broad range of universal Optical character Recognition (OCR) features and capabilities. Outbound mail stream processing is limited to reading the country name on mail leaving Germany. Therefore, OCR locates and reads a country name, which may be written in German as well as in the language of the destination country. For example, a letter addressed to the United States may have the country name written as "USA" or "Vereinigte Staaten." A letter addressed to the Ivory Coast may have "Côte d'Ivoire" or "Elfenbeinküste." It is important that a postal sorting machine be able to correctly interpret any variation, in writing, of the country name.
Spam Detection: Those who are using Email since 1900 may know this change in email clients, Since early 2000 we have started seeing a feature in mail box that automatically transfer unrelated and suspicious mail to junk folder. Earlier these systems were little faulty but now days these systems are far better.
Credit Card Fraud Detection: now days when you apply for credit card, the decision wheather or not to approve that card to you depends on your past purchase history, your past transaction and many other factors. A model is there to make this decision. Even Now days such machine leaning models also decides what credit amount should be approved to particular customer.
Speech Understanding: You are witnessing this change noways. Siri, Cortana, jarvis and google voice all most all tools have some talent of being able to make human like communication. This is machine learning right now in your hand.

These are few example to say, there are many more examples in field of Customer Segmentation, Stock Trading, Medical Diagnosis and image & video processing. In above said example no one made explicite programming for any of the tasks.

Explicit programming for all of the above given task is almost impossible. Lets take example of spam filtering; can anyone make all rules by if else that weather or not to classify given email as spams? answer is No. There can be millions of type of email and no buddy can make rules for all. Similarly no buddy can make rules for millions of faces on facebook.

Now we have some idea about machine learning, lets see why machine learning is a "buzz word" now.

Machine learning techniques exist since 1960, but were not in use much because of following reason.

Earlier algorithm were not powerful

Although all form of modern neural networks exist since 1960 s but, were not that powerful
in 2006 Geoffrey Hinton coins the term “deep learning” to explain new algorithms that let computers “see” and distinguish objects and text in images and videos. This is derived form of neural network but much more powerful.
Technically I need to explain two term/challenges those were there with earlier algorithm :
Vanishing gradients: Earlier network were not able to communicate/correct their error properly.
Overfitting : model made from particular data used to perform better on that particular data-set only, but used to show degraded performance.

Computational Need / Moore's Law

As said by Moore's says computational power doubles every 18 months. Earlier computers were not that powerful to handle memory and computational requirement by machine learning algorithm. Development in machine learning is incremental in the aspect of size of data and pattern recognition. But the curse of dimensionality is one big problem that is difficult to handle in machine learning. Now with increased computational power we can handle the curse of dimensionality

Digital Drift

Now days number of digital device exist on the earth are more than human population. 2.5 QUINTILLION BYTES of data is generated by these devices daily. Machine learning , precisely Deep Learning techniques are data hungry, more the data you feed more will be generalized intelligent and powerful model will be.
This is the reason why all companies are running behind data including Facebook, Google, Yahoo by providing free services like chat messengers, free photo storage services etc..

With this I complete, the section, "What is machine learning", lets move to "Why machine learning"

In this section I will present you with few facts that totally indicate that why machine learning is required.

So here I' m going to present data from report entitled as "McKinsey's 2016 Analytics Study Defines The Future Of Machine Learning"

Figure 1. Use of machine learning for various industries

Impact on Location based services:

All shopping, cab services, food delivery, logistics, use and sell, rent/lease etc are classified are location based services. In U.S. there is 60% rise in data and analytics used by location based services. location based services use machine learning for growing opportunities for businesses to use geo-spatial data to track assets, teams, and customers across dispersed locations to generate new insights and improve efficiency. U.S. Retail is capturing up to 40%, and Manufacturing, 30%.

forecasting and predictive analytics

Analysis was done across 120 industry and found that Machine Learning is really required for better production and performance. Data is as presented in heat map.

Broader impact of machine learning for various industries is as summarized below. The size of the bubble reflects the diversity of the available data sources.

Figure 2. Broader impact of machine learning for various industries

There much more to say and see in future about machine learning. One thing is very sure that this field is about to bring a revolution.

We have seen the most required, lets discuss "Why one need to learn Machine Learning"

Money matters to all of us, as a first factor of job satisfaction. As per the survey published by Stackoverflow for 2016, data scientist (Machine Learning) ramain one of the highest paying job with very less people already expert in this area (less competition)