Neural network simulation (character recognition) using mathematica / Khairul Anuar Muhammmad
In this project Backpropogation technique has been chosen to train data and to test the data. This technique is selected because it is the most common technique in Artificiai Neural Network simuïation. The studies that had been carried out ïn this project is to simulate neural network using BPN (Bac...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
1998
|
| Online Access: | https://ir.uitm.edu.my/id/eprint/103198/1/103198.pdf https://ir.uitm.edu.my/id/eprint/103198/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In this project Backpropogation technique has been chosen to train data and to test the data. This technique is selected because it is the most common technique in Artificiai Neural Network simuïation. The studies that had been carried out ïn this project is to simulate neural network using BPN (Backpropagation network) to recognize the capital letters and numbers. The BPN is a iayered, feedforward that is fully interconnected by layers. There is no feedback connections and no connections that bypass one tayer to go directly to a later ïayer. Because it is so powerful, the backpropagation network has become an industry Standard. Among the advantages of backprop are its abiiity to store rmmbers of pattems far in excess of its built-in vector dimensionality. The network sometimes may fait when trying to solve real problems, where it fail to converge after a large number of training set. When this phenomenon occurs, changes has to be made by adjusting the weight initiahzation, learning rate, and adding extra parameter such as momentum. |
|---|