A COMPARISON OF CONSTRUCTIVE AND PRUNING ALGORITHMS TO DESIGN NEURAL NETWORKS
Journal Title: Indian Journal of Computer Science and Engineering - Year 2011, Vol 2, Issue 3
Abstract
This paper presents a comparison between constructive and pruning algorithms to design Neural Network (NN). Both algorithms have advantages as well as drawbacks while designing the architecture of NN. Constructive algorithm is computationally economic because it simply specifies straightforward initial NN architecture. Whereas the large initial NN size of pruning algorithm allows reasonably quick learning with reduced complexity. Two popular ideas from two categories: “cascade-correlation [1]” from constructive algorithms and “skeletonization [2]” from pruning algorithms are chosen here. They have been tested on several benchmark problems in machine learning and NNs. These are the cancer, the credit card, the heart disease, the thyroid and the soybean problems. The simulation results show the number of iterations during the training period and the generalization ability of NNs designed by using these algorithms for these problems.
Authors and Affiliations
KAZI MD. ROKIBUL ALAM , BIKASH CHANDRA KARMOKAR , MD. KIBRIA SIDDIQUEE
Keywords
Related Articles
- EP ID EP129556
- DOI -
- Views 148
- Downloads 0
SPEECH ENHANCEMENT USING PARTICLE FILTERS:A CRITICAL REVIEW
Speech Enhancement refers to improvement in quality of a degraded speech signal in broad sense, however the aim is not only improvement of intelligibility but also overall quality, widely applied in de-reverberation, voi...
GIRD COMPUTING- A TOOL FOR ENHANCING THE COMPUTING POWER
With the enormous increase in the demand for computing capacities, solutions with least investment have to found out. In this direction Grid technology is finding its way out of the academic incubator and entering into c...
ENERGY AWARE ROUTING ALGORITHM FOR WIRELESS SENSOR NETWORK
Technological advancements made major research challenges in the area of wireless sensor network. Power saving is the foremost criteria in this field. Consumption of energy is considerably reduced on properly designed pr...
VARIABLE RATE STEGANOGRAPH Y IN DIGITAL IMAGES USING TWO, THREE AND FOUR NEIGHBOR PIXELS
In this paper three image steganography methods with variable rate of embedding are proposed. In the first method, called two neighbor method the two neighboring pixels such as upper and left are exploited to take embedd...
ANALYSIS ON PERFORMANCE WITH COMBINATION OF SCHEDULING AND REPLICATION IN DATA GRIDS
Data grid is the storage component of a grid environment. Amount of data transferred among nodes can be reduced by submitting the jobs to the nodes that having maximum requested files by scheduling and reducing the acces...