There is always a simple way of explaining difficult concepts and ideas. Attaining the simple way is not an easy task but it does make understanding easier for those who are interested. So, we will make an attempt here.

What is machine learning?

It is the process of teaching a machine to learn. Now, let us simplify that further. Imagine a child that has never seen or experienced fire sees a burning candle for the first time. He crawls up to it and touches the flame. The instant burn teaches him never to touch a flame ever again. The child learnt it by himself and he will grow up to apply the knowledge for any flames – burning lava, a gas burner, fireplace – he encounters.

Now, that is how we human beings learn by ourselves, through rewards or penalties, through exposure to new experiences, through the knowledge from our elders, etc. It is pretty much the same for machines. Only we dictate the purpose of their learning.

Now, let us classify.

Supervised learning

In this case the programme is trained with labeled data where both the input and the likely outputs are predetermined. After the algorithms are trained, they are exposed to unforeseen data which they have to classify.

Predicting the weather on a certain day based on earlier weather patterns, identifying a potential loan defaulter based on previous actions, these are instances of supervised learning problems. This form of machine learning is used for classification and regression models.

Unsupervised learning

As the name suggests, unsupervised learning models are trained with unlabeled data. These are used to cluster data based on similarities or dissimilarities. These models look for certain predetermined features in data, and often identify new features, based on the algorithm. This form is used for clustering.

Reinforcement learning

Reinforcement learning is based on a policy of rewarding correct prediction and penalizing wrong prediction. In this case the model reaches predictive accuracy through multiple trials and errors where each imperfection in results is penalized.

What is a neural network?

An artificial neural network is a close emulation of the animal nervous system used for locating complex patterns in data sets. A neural network consists of numerous layers or stacks of nodes or neurons. These can be input layers if they face the input data, output layers, and hidden layers.

Deep learning

The processing power of a neural network depends on its depth or the number of layers it features. As the complexity of data increases day by day, the neural networks require more depth. The machine learning process that uses a large number of layers or a deep neural network is termed deep learning.

Deep learning uses convolutional neural networks (CNN) for computer vision or image processing, and recurrent neural networks (RNN) for natural language understanding.

A business case for deep learning

Technological advancements find meaning through their use cases. Any technology that lacks a business case hardly gets out of the academic closet. The concepts of neural networks and deep learning are not new. Scientists have been working on these for over half a century. Thanks to the availability of data and the massive enhancements in computational power, deep learning has multiple serious business cases, hence, the popularity.

Manufacturers are using machine learning to manage their supply chain. There is a wide usage of computer vision at manufacturing units for quality control. The healthcare sector uses computer vision for analyzing radiological images. Pharmaceutical companies are using deep learning to analyze protein patterns. Banks and financial institutes have important use cases of sentiment analysis and natural language processing. The deep learning industry is juiced up and will stay like that for the foreseeable future.

A career in machine learning

If you are in India, nothing can beat a machine learning course in Bangalore to get you started. A strong base in statistics, grasp over one or two scripting languages, understanding of the mathematical base of algorithms, and exposition to a lot of previously done work, these are aspects that you need to focus on to get a strong foothold in machine learning.

Make it a habit to read about the field of your interest regularly. Plain awareness often makes the difference between success and failure. Start reading about how to learn machine learning on your own, but keep an eye open for any opportunity of learning from industry experts.

Learn about the domain that you are going to work in, understand its challenges, and goals, and keep those in mind while prescribing or implementing your solutions.