EMAIL: PASSWORD:
Front Office
UPT. PERPUSTAKAAN
Institut Teknologi Sepuluh Nopember Surabaya


Kampus ITS Sukolilo - Surabaya 60111

Phone : 031-5921733 , 5923623
Fax : 031-5937774
E-mail : libits@its.ac.id
Website : http://library.its.ac.id

Support (Customer Service) :
timit_perpus@its.ac.id




Welcome..guys!

Have a problem with your access?
Please, contact our technical support below:
LIVE SUPPORT


Moh. Fandika Aqsa


Davi Wahyuni


Tondo Indra Nyata


Anis Wulandari


Ansi Aflacha




ITS » Undergraduate Theses » Teknik Informatika
Posted by ida at 09/12/2009 16:11:12  •  4267 Views


PENINGKATAN PERFORMA MEMORI ASOSIATIF UNTUK PEMBELAJARAN POLA BIPOLAR DAN NONBIPOLAR MENGGUNAKAN JARINGAN SARAF TIRUAN REKUREN NONLINIER

PERFORMANCE IMPROVEMENT OF ASSOCIATIVE MEMORY FOR LEARNING BIPOLAR AND NONBIPOLAR PATTERNS USING RECURRENT NONLINEAR ARTIFICIAL NEURAL NETWORK

Author :
Destyanto, Rahaditya ( 5103109066 )




ABSTRAK

Masalah utama dalam pembuatan suatu model memori asosiatif adalah maksimasi kapasitas memori dan peningkatan tingkat temu kembali. Selain model Hopfield yang sederhana telah dikenalkan beberapa model yang memiliki kapasitas memori yang lebih besar dan kemampuan untuk mempelajari pola berkorelasi yang lebih baik dengan menggunakan aproksimasi terhadap aturan pseudo-inverse. Namun modelmodel tersebut masih menghasilkan jumlah spurious attractor cukup banyak sehingga daya temu kembali memori tersebut rendah. Model-model tersebut juga memerlukan pre-processing untuk mempelajari pola-pola nonbipolar. Model baru yang menggunakan fungsi nonlinier untuk pembelajaran dan recall diduga mampu menghasilkan lebih sedikit spurious attractor sehingga memiliki tingkat temu kembali yang tinggi. Model ini juga mampu membentuk attractor bernilai real sehingga dapat memproses pola-pola nonbipolar tanpa preprocessing. Performa model memori asosiatif baru tersebut diamati dalam tugas akhir ini dengan mempelajari pola-pola bipolar berupa gambar hitam-putih dan pola-pola nonbipolar berupa gambar dengan tingkat keabuan 16 warna. Hasil yang didapatkan menunjukkan performa temu kembali yang lebih baik dibandingkan model-model sebelumnya yaitu v 5946 untuk tingkat noise real sebesar 200 dibandingkan dengan performa terbaik model-model sebelumnya yaitu sekitar 17. Untuk input dengan inversi pixel performa yang diberikan adalah 69 dibandingkan dengan performa memori linear sebesar 63. Memori asosiatif nonlinier ini juga dapat mempelajari pola-pola nonbipolar tanpa pre-processing yang ditunjukkan dengan keberhasilan melengkapi pola dengan tingkat kerusakan 50.


ABSTRACT

The main problem of associative memory models construction is the maximization of memory capacity and recall performance. New improved models have been proposed that offer bigger memory capacity and better recall performance of correlated patterns than the simple Hopfield model based on approximation of the pseudo-inverse rule. However the models still develop numerous spurious attractors causing poor recall performances. Those models also need pre-processing to learn nonbipolar patterns. The new proposed model which uses nonlinear activation function for learning and recall is suspected to be able to develop less spurious attractors therefore have high recall performance. The model differs from previous models for its ability to develop real value attractors as a result the model needs no preprocessing to learn nonbipolar patterns. Performance of the new model is observed in this final project by learning bipolar patterns in the form of black and white images as well as nonbipolar patterns in the form of 16- depth grayscale images. Results show that the model is able to develop less spurious attractors and have better recall performance than the previous models. Successful recall rate under noise proportion of 200 is vii 59.46 while the next best performance by previous model is approximately 17. For input with flipped pixels the new model is able to recall 69 noisy inputs compared to 63 success rate of linear models. The model is also able to learn nonbipolar patterns without any pre-processing by successfully recall incomplete patterns of 50 loss rate.



Keywordsassociative memory, artificial neural network, bipolar pattern, nonbipolar pattern.
 
Subject:  Perangkat lunak komputer ; Jaringan syaraf tiruan
Contributor
  1. Rully Soelaiman, S.Kom., M.Kom.
    Mediana Aryuni, S.Kom., M.Kom.
Date Create: 09/12/2009
Type: Text
Format: pdf
Language: Indonesian
Identifier: ITS-Undergraduate-3100007028699
Collection ID: 3100007028699
Call Number: RSIf 006.32 Des p


Source
Undergraduate Theses of DEPARTEMENT OF INFORMATICS ENGINEERING Faculty of Information Technology, RSIf 006.32 Des p, 2007

Coverage
ITS Community Only

Rights
Copyright @2007 by ITS Library. This p ublication is protected by copyright and permission should be obtained from the ITS Library prior to any prohibited reproduction, storage in a retrievel system, or transmission in any form or by any means, electronic, mechanical, photocopying, recording, or likewise. For information regarding permission(s), write to ITS Library




[ Download - Summary ]

ITS-Undergraduate-3100007028699-5922.pdf




 Similar Document...




! ATTENTION !

To facilitate the activation process, please fill out the member application form correctly and completely

Registration activation of our members will process up to max 24 hours (confirm by email). Please wait patiently

POLLING

Bagaimana pendapat Anda tentang layanan repository kami ?

Bagus Sekali
Baik
Biasa
Jelek
Mengecewakan





You are connected from 3.235.85.115
using CCBot/2.0 (https://commoncrawl.org/faq/)



Copyright © ITS Library 2006 - 2020 - All rights reserved.
Dublin Core Metadata Initiative and OpenArchives Compatible
Developed by Hassan