Machine learning (ML) is one of the most trending topics of Artificial Intelligence (AI) which, though not new in concept is becoming increasingly popular day by day. Today, Machine learning is being used in numerous Computer Science, Artificial Intelligence and IoT applications.

Many among us are still not properly aware of this concept and its necessity even while using it. Strictly speaking, Machine learning is a subset of Artificial intelligence which is driving the most fascinating and disruptive technologies.

AI is playing major role in the following fields:

• Natural Language Processing.
• Speech.
• Computer vision.
• Robotics.
• Expert systems.
• Planning, scheduling and optimization.

A recent report by the Future of Humanity Institute presents a survey from a panel of AI researchers on timelines for Artificial General Intelligence (AGI), and the report stated that:

“Researchers believe there is a 50% chance of AI outperforming humans in all tasks in the next 45 years” (Grace et al, 2017)

This will automate all human jobs in the next 120 years. All of these examples are not just hypotheses or imagination. As AI is the future of technology so is Machine Learning.

Now keeping our discussion limited to Machine Learning today, let’s start with its formal definition.

“Making a machine to learn and take decisions just like a normal human.”

Though it sounds cool but this is nothing new. It’s just an application of all the statistics and linear algebra which we have studied and apply in all major aspects of our daily life, without even being aware of it. “Machine learning” is a way of “training” an algorithm so that it can learn to predict or to classify or to categorize things. Training involves feeding huge amounts of data to the algorithm and allowing the algorithm to adjust and improve on its own.

Human and Machine

This comparison can be related to the most common example which every individual must have observed in their daily life i.e. our elders consciously or subconsciously take more accurate and precise decisions than us due to their prior knowledge and experience. Is it like that same situation happened with them before? No, similar situation might have happened to them or they might have observed it in their surroundings.

This leads the foundation of “learning”. Just as we study psychology and neuroscience to understand how humans learn, decide, act, and feel. In the similar way we can train machine to predict, analyses, classify data and take decisions. We can build machine models that can perform well (close to humans) to some extent.

Social Media and Machine Learning

Social media sites Facebook, Twitter, Instagram, and LinkedIn are using Machine Learning in their systems. From suggesting people you may know to automatically tagging you in photos of you and displaying news feed of your interests, everywhere we see the use of Machine Learning.

The science behind this is that they are actually analyzing your data which includes your interests, history and pictures, and program their algorithms using the data obtained from user activities.

Online Business/E-commerce and Machine Learning

Every online business aims to improve its search engine presence. For this purpose, you can use recommendation systems. These are Machine Learning algorithms that help in recommending relevant items to the user. Data set of the online system gets updated according to the activity of users.

It actually filters the search of an individual user based on their likes and dislikes. So, instead of hard coding software routines with specific instructions to accomplish a particular task, you will just need a set of Machine learning models for training.

Some popular examples are of Amazon (which often recommends books to buy) and Netflix (which recommends movies to rent). The system used by these organizations learns what sort of items you would like to buy or what sort of movies you would like to watch, and can therefore give customized interest based recommendations to you.

Basics of Machine Learning

If you are new to machine learning you need to understand these two types of Machine learning. Supervised learning and unsupervised learning.

Supervised Learning

It is the learning which is based on training of data that is already labelled, i.e. in this type of learning we have labels for each instance/example/observation of data. There are further sub-types of supervised learning, and they are:

Predictions/Regression: We want to predict something or the outcomes on the basis of some input. For example, if we want to predict the stock exchange rates or we want to predict the price of any house /we can apply regression techniques of ML.

Classification: We want to classify our data into some category; and there can be two classes (binary classification) or more than two classes (multi-class classification). For example, identifying spam and fraud detection.

Unsupervised Learning

Segmentation clustering groups when the data set is without labels we have to move from supervised learning to unsupervised learning. Basically, we do not know anything in advance.

However, on the basis of characteristics and behavior we can classify our data set into clusters and groups. In unsupervised learning we find some patterns and relations in data set. For example, when we want to find the segmentation of market while opening a new business we can target maximum audience. This can be done with the help of unsupervised learning.

When should we use Machine Learning?

Suppose you have a situation and you don’t exactly know is it ML based problem or can you optimize it using ML.

Then what you need to look on?

1. Do I need this machine learning solution in future as well?

2. Will the size of my data vary?

3. Do I require customization in the solution?

4. Do I need to scale my product?

If the answer of all the above questions is yes then your problem can be optimized with Machine learning.

“Machine learning provides systems the ability to automatically learn and improve from experience without being explicitly programmed”

Let me explain explicitly what being programmed means. It means you do not need to write million lines of code for each condition or each logic which for sure is not a smart and efficient technique. Instead, you just need to train the machine to find the relation between the data sets and then take decisions smartly.

Limitations

1. Not every model can be applied on every data set. You need to find the perfect model for your data set.
2. You need a huge data set to program your algorithm for training.

Conclusion

Understanding Machine Learning at a conceptual level is definitely worth it. It will bring revolution in the industry while having massive impact on our day to day routine work as it will allow us to filter the material we read and the tools we choose in order to stay focused on the problems we are trying to solve.