Abstract:Training time of traditional multilayer perceptrons (MLPs) using back-propagation algorithm rises seriously with the problem scale. For multi-class problems, the convergence ratio is very low for training MLPs. The huge time-consuming and low convergence ratio greatly restricts the applications of MLPs on problems with tens and thousands of samples. To deal with these disadvantages, this paper proposes a fast BP network with dynamic sample selection (BPNDSS) method which can dynamically select the samples containing more contribution to the variation of the decision boundary for training after each iteration epoch. The proposed BPNDSS can significantly increase the training speed by only selecting a small subset of the whole samples. Moreover, two kinds of modular single-hidden-layer approaches are adopted to decompose a multi-class problem into multiple binary-class sub-problems, which result in the high rate of convergence. The experiments on Letter and MNIST handwritten recognition database show the effectiveness and the efficiency of BPNDSS. Moreover, BPNDSS results in comparable classification performance to the convolutional neural networks (CNNs), support vector machine, Adaboost, C4.5, and nearest neighbour algorithms. To further demonstrate the training speed improvement of the dynamic sample selection approach on large-scale datasets, we modify CNN to propose a dynamic sample selection CNN (DynCNN). Experiments on Image-Net dataset illustrate that DynCNN can result in similar performance to CNN, but consume less training time.

DSP Fast Processing Technology Based on BP Algorithm

Research on Data Acquisition Algorithms Based on Image Processing and Artificial Intelligence

Realization and Optimization of Face Detection Algorithm Based on DSP

Study on Preprocessing Algorithm of Printed Symbols

3D Image Recognition Technology for Analog Circuit Based on DSP and BP Algorithm

Adaptive preprocessing of character recognition image based on neural network

Fast recognition method for printed numbers

Research on Counting Methods of Deposits Based on DSP

Realization of Fingerprint Identification on DSP

A High Speed Color Face Pre - processing Algorithm in Complex Background

A Fast BP Networks with Dynamic Sample Selection for Handwritten Recognition.

Character Recognition Based on Improved BP Neural Network

Research on Parallelization of Face Recognition

Fast Processing of UAV Remote Sensing Image Based on DSP Hardware Simulation Platform

A Versatile High-speed Image Processing System Based on DSP and CPLD

A New Fast Image Encoding Algorithm for DSP

Design and Application of Fire Images Recognition System Based on DSP

Bistatic PFA Parallel Algorithm Based on Double Chirp-Z Transforms and the Dsp Implementation

Image processing methods for visual prostheses based on DSP

Digit Character Recognition Based on Wiener Filter, Karhunen-Loeve Transform and BP Network

Preprocessing Technique in Fingerprint Recognition System