Design and evaluation of neural networks for an embedded application

Authors:
Paolo Motto Ros;Eros Pasero
Affiliations:
Laboratorio di Neuronica, Dipartimento di Elettronica, Politecnico di Torino, Torino, Italy;Laboratorio di Neuronica, Dipartimento di Elettronica, Politecnico di Torino, Torino, Italy
Venue:
IEA/AIE'10 Proceedings of the 23rd international conference on Industrial engineering and other applications of applied intelligent systems - Volume Part III
Year:
2010

Citing 6
Cited 1

Technical Note: Selecting a Classification Method by Cross-Validation

Machine Learning
On the capabilities of neural networks using limited precision weights

Neural Networks
Introduction to Data Mining, (First Edition)

Introduction to Data Mining, (First Edition)
Learning from Imbalanced Data

IEEE Transactions on Knowledge and Data Engineering
A study of cross-validation and bootstrap for accuracy estimation and model selection

IJCAI'95 Proceedings of the 14th international joint conference on Artificial intelligence - Volume 2
An overview of statistical learning theory

IEEE Transactions on Neural Networks

A Comparison Study for a Neural Network Based Embedded Appliance

Proceedings of the 2011 conference on Neural Nets WIRN10: Proceedings of the 20th Italian Workshop on Neural Nets

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this study we compare different training strategies and configurations of Artificial Neural Networks (ANNs) designed for a real application, a mobile reader device for blind people. They are used to recognize characters found in input images; since the application has a real-time behaviour, we have improved the already implemented ANN subsystem rather than looking for new and more complex solutions. The main idea has been to setup a pool of configurations and then to select the best one through a rigorous test. All the results have been computed on both the development platform and the target one, considering implementation issues. This method has led to improve the performances of the classifier maintaining the same complexity as the starting one.