Artificial Intelligence is the science of development of computer systems that can mimic human actions like perception, speech recognition, language processing and decision making. Artificial Intelligence algorithms had been conceived by people way before the second industrial revolution i.e. the computing age. It is a rapidly growing interdisciplinary domain that has seen the participation of researchers from Psychology, Electrical Engineering, Computer Science, Economics, Biotechnology and a host of other fields.
After the landmark paper of Rosenblatt in 1957 that introduced the Perceptron machine (Algorithm) , it was widely felt in the scientific community that the ideas and concepts which were scattered and unimplementable could finally be realised in practice. However, within a few years, people realised the Perceptron Algorithm was not as amazing as had been advertised by its creators. Minsky and Papert’s paper squashing the applicability of Perceptron and rendering it useless only except at special situations led to a new low in AI research. A dream which was thought to have been almost translated to reality had been thrust back into the inner recesses of the imaginative mind.
It was in the beginning of the 21st century that history was created in Artificial Intelligence by a person who was primarily a Cognitive Psychologist. His name – Geoffrey Hinton. Hinton and team made a revolutionizing discovery in what is now known as Deep Learning. They demonstrated that Perceptrons with many layers (known as a multi layer neural network) when empowered with the Back propagation algorithm can do wonders in solving many of the traditional problems like Classification and Regression. The study of neural networks has seen a rapid rise of interest ever since and current state of the art neural nets are responsible for all the computing miracles we see around us. Siri, Cortana, Pokemon Go and many other fancy softwares manufactured these days have a neural net at the heart of their functioning.
Learning is the most important paradigm of Artificial Intelligence. Training a machine to learn to identify objects, replicate speech, perform motion are some of the things that Machine learning achieves. Machine learning is the part of AI that ”works”. Various probabilistic and deterministic tools are used to create machine learning algorithms. Some of these include Support Vector Machines, Fuzzy Logic, Logistic Regression , Expectation Maximization procedures and Artificial Neural Networks(Nets for short). After the beginning of the Deep Learning era, Neural Nets are currently the hottest tools being used in research and implementation.
Artificial Neural Networks attempt to mimic the human learning (thinking) process by making an artificial model of the human nervous system. Every neural net has at least three layers, namely the Input layer, Hidden layer(s) and the Output layer. Deep learning is a current state of the art method used to model higher level abstractions in data by using multiple hidden layers in neural nets. The individual processing units in a neural network are called ’neurons’ which perform arithmetic operations on the input data and send an output to the next layer of neurons. The neurons of every layer are interconnected in a complicated way to extract ’features’ out of the input data. At every step, higher level ’features’ are extracted and passed on to the next layer. The ex- act mechanism of what is ’learnt’ and how the learning occurs in an Artificial Neural Network is still unknown, but what is known is that it works pretty well in a variety of tasks like Computer Vision , Natural Language Processing and Speech Recognition. They have been used for making Restricted Boltzman machines(RBM), Convolutional Neural Networks(CNN), Long Short Term Memory Networks (LSTM) which are currently used in myriad software applications. These tools are used primarily as ’black boxes’ for use in domains different from Electrical Engineering and Computer Science.
Millions of dollars are currently being invested by big industrial giants like Google and Facebook to expand research in Deep Learning. Looking at the growing interest of academicians and Industrialists in exploiting Deep Learning and developing its applications, it is safe to assume that this is the next big thing in the world of innovation.