City Research Online

Fat-Fast VG-RAM WNN: A high performance approach

Forechi, A., De Souza, A.F., Neto, J.D.O. , de Aguiar, E., Badue, C., Garcez, A. & Oliveira-Santos, T. (2016). Fat-Fast VG-RAM WNN: A high performance approach. NEUROCOMPUTING, 183, pp. 56-69. doi: 10.1016/j.neucom.2015.06.104

Abstract

The Virtual Generalizing Random Access Memory Weightless Neural Network (VGRAM WNN) is a type of WNN that only requires storage capacity proportional to the training set. As such, it is an effective machine learning technique that offers simple implementation and fast training – it can be made in one shot. However, the VG-RAM WNN test time for applications that require many training samples can be large, since it increases with the size of the memory of each neuron. In this paper, we present Fat-Fast VG-RAM WNNs. Fat-Fast VG-RAM WNNs employ multi-index chained hashing for fast neuron memory search. Our chained hashing technique increases the VG-RAM memory consumption (fat) but reduces test time substantially (fast), while keeping most of its machine learning performance. To address the memory consumption problem, we employ a data clustering technique to reduce the overall size of the neurons’ memory. This can be achieved by replacing clusters of neurons’ memory by their respective centroid values. With our approach, we were able to reduce VG-RAM WNN test time and memory footprint, while maintaining a high and acceptable machine learning performance. We performed experiments with the Fat-Fast VG-RAM WNN applied to two recognition problems: (i) handwritten digit recognition, and (ii) traffic sign recognition. Our experimental results showed that, in both recognition problems, our new VG-RAM WNN approach was able to run three orders of magnitude faster and consume two orders of magnitude less memory than standard VG-RAM, while
experiencing only a small reduction in recognition performance.

Publication Type: Article
Publisher Keywords: WNN, VG-RAM, multi-index chained hashing, data clustering, traffic sign recognition, handwritten digit recognition.
Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Departments: School of Science & Technology > Computer Science
SWORD Depositor:
[thumbnail of FatFast.pdf]
Preview
Text - Accepted Version
Available under License Creative Commons: Attribution 3.0.

Download (1MB) | Preview

Export

Add to AnyAdd to TwitterAdd to FacebookAdd to LinkedinAdd to PinterestAdd to Email

Downloads

Downloads per month over past year

View more statistics

Actions (login required)

Admin Login Admin Login