Introduction to Deep Learning

January 2, 2018

Introduction

With tremendous advancement in technology, having intelligent machines is no surprise. The machines available today are ‘smart’ that can make basic decisions without human supervision and thus help make our lives easier. Technology that makes these machines smart is called ‘Artificial Intelligence’ wherein a machine is fed with thousands of lines of coding dictating its actions in different scenarios.



Artificial Intelligence, though a great technology, is very complex and often expensive. It involves extensive coding to get a machine to do simple tasks. To overcome this, concept of ‘Machine Learning’ was brought in. In Machine Learning, a machine is trained specifically for a task. It is fed with enormous amount of data involving the task that it is expected to accomplish. Eventually, machines learn how to do that task by forming its own set of algorithms. Introduction of machine learning did complete the task of effectively replacing Artificial Intelligence in areas where it was not required, however, making the customer experience better was one scope that could be explored further.

What is Deep Learning?
‘Deep Learning’ is a subset of Machine Learning that uses multiple layers of artificial neurons for classification and pattern recognition. It works on the principle of identifying the preferences of the user and then showing result accordingly. Ever wondered how the keyboard of your smart phone learns the words that you frequently use, even if they are grammatically incorrect? Or how automatically Facebook finds and tags your friends in a photograph? These are both examples of Deep Learning. Laptops and phones learn preferences and likes of a user using this concept and then make the user experience better with this knowledge.

Chat Bots
The most recent and prominent example of deep learning is available to us in the form of chat bots. Chat bots are deployed by many websites today to provide their users a platform for connecting with the website administrator. A Bot is nothing but a deep learning algorithm that is initially trained according to the questions frequently asked by the website visitors. Chat bot is a self-learning and self-improving algorithm that gets better with time and use. Apart from this, there are many places where deep learning algorithms can be found.
Speech recognition – It is fascinating how Siri and Cortana understand our speech and convert it to text while executing our command. Speech recognition is a function of deep learning where the algorithm is designed to sample a speech signal and then process it to determine the text form of it.

Automatic machine translation – In an era where boundaries hold no physical meaning, language shouldn’t be a barrier either. Automatic text translation is a function that translates a text to any chosen language. This function is achieved through deep learning algorithms trained with multiple languages.

Instant visual text translation – Suppose you were in a country where the native language was unknown to you, understanding their texts would be an impossible task. However, with deep learning algorithms, clicking a picture of foreign text and getting it translated to your native language is now possible. This feature of deep learning is an advancement of its property of object recognition.

Self-driving cars – This concept is still new where a car is programmed in such a way that it can drive itself around the city. Navigating through traffic, understanding routes and understanding of pedestrians are aspects of human intelligence. Its incorporation in a car’s working has been made possible with deep learning algorithms
Future applications

Glasses for blind – Using the deep learning algorithms, a device can be trained to recognise things around itself and convert it to audio. For example, a camera backed up with a software that captures its surroundings (a tree, a bicycle, a kid, a turn, etc.) and recognises the image to converts it to an audio. This feature can be used to make glasses for blind people that could replace their need of being navigated to places by sticks and humans.

Personalised medicines – Every individual in this world is different from the other and hence everybody reacts to a medicine differently. Using deep learning, a person could be studied with a personalised algorithm enabling the doctors to design a specific medicine for the patient that would suit him completely.

Taking out match highlights – In the current scenario, every interesting moment of a match is determined by the human supervisor sitting behind the camera. With deep learning in picture, designing an algorithm that has an understanding of the interesting moments of a match is possible. The highlights could then be taken out without human supervision that would eventually lead to lesser errors.
Created with