Comparison And Analysis Of Perceptron And BP Algorithms

Perceptrons algorithm is a kind of effective online algorithm which has the ability to classify linearly separable training patterns correctly. Furthermore, understanding the method of perceptrons algorithm can better lay the foundation for us to understand other complex neural network models. BP algorithm is widely used in many applications.It has ability to classify XOR problems, but it has some problems for example it always sinks into local minima and has a slow speed.In this paper, according to the characteristics of perceptron algorithm and BP algorithm, we compare their effectiveness of classification and the speed of training for linearly separable and linearly non-separable problems.In the past the comparison of perceptron and BP have adopted their own standards, we unify the convergence condition and the counting of training steps. We attempt to take a new look at the applicability of perceptron.We introduce the parallel perceptron, and propose that XOR problem with n-dimension can be classified by parallel perceptron if and only if the hidden neuron's number of the parallel perceptron is larger and equal than n.We also compare the effectiveness of parallel perceptron, MR I and BP.The structure of this thesis is organized as follows:Chapter 1 reviews the history and basic knowledge of neural networks, also gives a brief introduction of neuron model,liner perceptrons and BP algorithm.Chapter 2 focuses on the study of liner perceptron and its relationship to linearly separable problems, and then discusses the convergence of perceptrons.In chapter 3,according to the characteristics of perceptron algorithm and BP algorithm, we compare their effectiveness of classification and the speed of training for linearly separable and linearly non-separable problems.Chapter 4 introduces the parallel perceptron.We also estimate parallel perceptron's classification capacity of XOR problem.