(For USM Staff/Student Only)

EngLib USM > Ω School of Electrical & Electronic Engineering >

A hybrid evolutionary neural network for pattern classification / Loo Chin Sing

A hybrid evolutionary neural network for pattern classification_Loo Chin Sing_E3_2010_875003648_00003084342_NI
Dalam tesis ini, satu evolusi buatan rangkaian saraf (evolutionary artificial neural network – EANN) yang novel telah dicadangkan, iaitu gabungan dari fuzzy ARTMAP (FAM) dan hibrida algoritma kekacauan-genetik (hybrid chaos-genetic algorithm - Hybrid CGA). Pengendali kacau pemetaan (chaotic mapping operator – CMO) yang berada di dalam CGA boleh menghasilkan satu populasi kromosom, yang mempunyai ciri pembolehubah kacau yang diedarkan dengan ketidakteraturan melalui seluruh ruangan carian. Hybrid CGA adalah sebuah dua-tahap hibrida algoritma carian yang novel, iaitu gabungan dari CGA dan carian langsung (direct search - DS). Hybrid CGA diyakini bahawa memiliki keuntungan yang sama dengan GA dan DS yang berada di Hybrid GA, tetapi tambahan dengen kemampuan carian global yang lebih tinggi. CGA pada pasa-1 boleh meningkatkan kemungkinan untuk mencari di ruangan global, dan kemudian menyediakan titik permulaan yang baik kepada fasa-2, dimana DS melakukan tempatan halus-penyesuaian untuk mendapatkan wilayah yang optima dengarn penyelesaian terbaik. Hybrid CGA yang dicadangkan kemudian digunakan untuk mengembangkan sambungan beban yang berada di dalam rangkaian FAM, Hybrid EANN yang terhasil dinamakan sebagai, FAM-Hybrid-CGA. Karakteristik kekacauan dalam CGA harus dapat meningkatkan kebarangkalian carian global yang diperolehi dengan FAM-Hybrid-CGA. FAM-Hybrid-CGA yang tercadang dipercayai bahawa mampu mengalahkan daripada perlawan (FAM-Hybrid-GA, iaitu yang tanpa CGA), dan bukti-bukti tentang keberkesanan ini akan dibentangkan dalam tesis ini. _____________________________________________________________________________________ In this thesis, a novel evolutionary artificial neural network (EANN) is proposed, that combines of fuzzy ARTMAP (FAM) and hybrid chaos-genetic algorithm (Hybrid CGA). The chaotic mapping operator (CMO) in CGA can generate a chromosome population, which having the characteristics of chaotic variable that irregularity distributed over the search place. The Hybrid CGA is a novel two-phase hybrid search algorithm, which are the combines of CGA and direct search (DS). The Hybrid CGA is believed that have the same advantages of GA and DS at Hybrid GA, but additional with the highest global search ability. The CGA at phase-1 can increased the chances of searching at global space, and then provide the good starting points to phase-2, where DS is perform local fine-tuning to get an optimum region with the best solutions. The proposed Hybrid CGA is then used to evolve the connection weights in FAM network, the resulted Hybrid EANN is so called, FAM-Hybrid-CGA. The chaos characteristic of CGA should enhance the global search probability of the FAM-Hybrid-CGA. The proposed FAM-Hybrid-CGA is believed that could outperform than its counterparts (FAM-Hybrid-GA, that without CGA), and the proof of its effectiveness are outlined in this thesis.
Contributor(s):
Loo, Chin Sing - Author
Primary Item Type:
Final Year Project
Identifiers:
Accession Number : 875003648
Barcode : 00003084342
Language:
English
Subject Keywords:
evolutionary ; hybrid chaos-genetic ; two-phase hybrid
First presented to the public:
1/4/2010
Original Publication Date:
3/15/2018
Previously Published By:
Universiti Sains Malaysia
Place Of Publication:
School of Electrical & Electronic Engineering
Citation:
Extents:
Number of Pages - 70
License Grantor / Date Granted:
  / ( View License )
Date Deposited
2018-03-15 15:54:58.962
Date Last Updated
2019-01-07 11:24:32.9118
Submitter:
Nor Hayati Ismail

All Versions

Thumbnail Name Version Created Date
A hybrid evolutionary neural network for pattern classification / Loo Chin Sing1 2018-03-15 15:54:58.962