Simant Dube - An Intuitive Exploration of Artificial Intelligence: Theory and Applications of Deep Learning

This Springer imprint is published by the registered company Springer Nature Switzerland AG

The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

I visualize a time when we will be to robots what dogs are to humans, and Im rooting for the machines.

Claude Shannon

Artificial intelligence would be the ultimate version of Google. The ultimate search engine that would understand everything on the web. It would understand exactly what you wanted, and it would give you the right thing. Were nowhere near doing that now. However, we can get incrementally closer to that, and that is basically what we work on.

Larry Page

In June 2014, we drove to the Computer History Museum in Mountain View, California, not too far from where I was living. A spectacular surprise was in store for us. As soon as we entered the building, our eyes were riveted on an 11 ft. (3.4 m) long and 7 ft. (2.1 m) high sculpture of engineering weighing five tonnes and having 8000 parts. One hundred and sixty-five years after it was originally conceived and designed, the Difference Engine 2 of Charles Babbage was operational in front of us. I wonder if the Cambridge mathematician from Victorian England could foresee that one day his work would be on display in the heart of Silicon Valley. We listened with rapt attention to the docent talk and demonstration of this marvelous machine.

I wish to God these calculations had been executed by steam! an anguished Charles Babbage exclaims as he finds error after error in astronomy tables calculated by human hand. It is London and the summer of 1821, and this statement is a remarkably prescient wish for an era of automation by machines that would carry out intelligent tasks without any errors. After 200 years, Larry Pages vision of the ultimate version of Google is not too different. Instead of the goal of automating tedious calculations, Google strives to build machines that will automatically understand everything on the World Wide Web.

What has changed in the intervening two centuries is the ubiquity of computing that has opened the doors to an unstoppable, colossal flood of data, also known as big data . Within the big data are numbers, observations, measurements, samples, metrics, pixels, voxels, 3-D range data, waveforms, words, symbols, fields, and recordswhich can all be processed and munged to expose structures and relationships. We want to use the big data to predict, classify, rank, decide, recommend, match, search, protect, analyze, visualize, understand, and reveal.

Of course, we are the creators of data, and we understand the underlying processes. We can bring our expertise to the fore and employ it to build machine learning models. This is the basis of classical ML, the parent of AI. Classical ML demands human labor during a process called feature engineering. It can work remarkably well for some applications, but it does have limits.

AI seeks to learn from the raw data. It chooses a large, deep neural network with tens or hundreds of millions of internal learnable parameters that are tweaked during training using calculus to minimize a suitably chosen loss function on a large annotated training dataset consisting of tens or hundreds of millions of examples. Once tweaking has been completed, the network can be deployed in production and used in the field to carry out the magical intelligent inference. During inference, AI models execute billions of multiplications, additions, and comparisons per second. If we were to create a mechanical slow-motion replica of such an AI model and witness a docent demonstration in a futuristic museum, it would truly be mind-boggling to see the countless moving parts of the endlessly sprawling machine as it crunched out a final answer.

This book is an endeavor to describe the science behind the making of such an impressive machine. One may askwhere is intelligence in the middle of billions of clinking, clanking parts? For human intelligence, we have first-hand subjective experience and we demand no further proof. For AI, the question is a difficult one. Let us take the first step of understanding how AI works.

This book is about developing conceptual understanding of Artificial Intelligence (AI), a rapidly moving field of growing complexity and of immense transformative potential. Deep learning has emerged as a key enabling component of AI. Significant attention is being paid to it by the media, governments, businesses, and the public. It can safely be said that AI will continue to remain an exciting frontier for humanity to seek and explore for centuries to come. Besides the motivation of using AI for its myriad practical applications, we humans like to contemplate and comprehend. This book seeks an understanding of AI and Machine Learning (ML) in the truest sense of the word. It is an earnest endeavor to unravel what is happening at the algorithmic level, grasp how applications are being built, and show the long adventurous road in the future. It is my ardent wish to answer some fundamental questions. Can we really create intelligent machines? Can an algorithm serve as a model of learning and intelligence? Will there always be some differences between human intelligence and AI? In order to answer these questions, the first step is to understand the state of the art in AI.