Neural Network and Back Propagation

Abstract

Humans and computers are suited to different types of tasks. Like, finding the sine of any number is difficult for humans, but easy for the computers. On the other hand, looking and identifying the image of a bird is very easy for us, as compared to the machines [1]. But nowadays, Artificial neural networks or ANNs are most commonly used in image processing. In fact, the recent deep learning algorithms have surpassed the human ability of image recognition. It is proved and stated that BPNN (Backpropagation Neural Networks) have a very high accuracy. A large part of recent success of neural network is explained on the basis of very high data availability. Thus, in B-P the network is trained with a lot of training examples so that it doesn’t compromise with the efficiency of the system. By the end of this century, it is expected that computers will be training neural networks with as many neurons as the human brain has [1][2]. Although it is difficult to predict what the true abilities of artificial intelligence will be by then, our experience with computer vision should prepare us to expect something unexpected. So, with this attempt paper tries to give the idea of how image is recognized and how back propagation works.