Newsgroups: comp.parallel
From: eskfoo@alpha.ntu.ac.sg (Hello... I'm Shou King)
Subject: ***** NEW BOOK ANOUNCEMENT *****
Organization: Nanyang Technological University - Singapore
Date: Tue, 16 Jul 1996 11:37:01 +0800
Message-ID: <4sintf$mpl@usenet.srv.cis.pitt.edu>

                             BOOK ANNOUNCEMENT
                             -----------------

Title:
        Parallel Implementations of Backpropagation Neural Networks on
        Transputers - A Study of Training Set Parallelism


Publication Date:

        July 1996


Brief description
-----------------

This book presents a systematic approach to parallel implementation of
feedforward neural networks on an array of transputers. The emphasis is
on backpropagation learning and training set parallelism. Using systematic
analysis, a theoretical model is  developed for the parallel implementation. 
The model is then used to find the optimal mapping to minimize
the training time for large backpropagation neural networks. The model
is also validated experimentally on several well known benchmark problems.
Use of genetic algorithms for optimizing the performance of the parallel
implementations is described. Guidelines for efficient parallel 
implementations are highlighted.

Table of contents
-----------------

1.      Introduction

        1.1 Multilayer Feedforward Neural Networks
        1.2 The Basic BP Algorithm
        1.3 Parallelism in the BP Algorithm
        1.4 Some Parallel Implementations

2.      Transputer Topologies for Parallel Implementation

        2.1 The Transputer
        2.2 Topologies
        2.3 Topology Chosen in this Study
        2.4 Software Used
        2.5 Performance Metrics and Benchmark Problems

3.      Development of a Theoretical Model for Training Set Parallelism
        in a Homogeneous Array of Transputers
     
        3.1 Time Components of Parallel Transputer Implementation
        3.2 Timing Aspects of Parallelizing the Backpropagation Algorithm
        3.3 Time Components for the Parallelized Backpropagation Algorithm
        3.4 Validation of the T-epoch  Model 

4.      Equal Distribution of Patterns Amongst a Homogeneous Array of
        Transputers

        4.1 Analytical Model for Time per Epoch
        4.2 Validation of the Model for Equal Distribution
        4.3 Optimal Number of Transputers Needed for the case of Equal
            distribution
        4.4 Cost Benefit Analysis of adding Additional Processors

5.      Optimization Model for Unequal Distribution of Patterns in a
        Homogeneous Array of Transputers

        5.1 Constraints for Optimization
        5.2 Optimal Pattern Distribution
        5.3 Validation of Pattern Optimization Model
        5.4 Experimental Results for Benchmark Problems
        5.5 Locating Surplus Processors and to Find out the Optimal
            number of Processors needed to Obtain Minimum Time per epoch

6.      Optimization Model for Unequal Distribution of Patterns in a
        Heterogeneous Array of Transputers
    
        6.1 Experimental Results for Benchmark Problems
        6.2 Statistical Verification of the Optimal Epoch Time
        6.3 Discussion

7.      Pattern Allocation Schemes Using Genetic Algorithm

        7.1 Optimization Algorithm and Computational Complexity
        7.2 Solution Time for Optimal Pattern Distribution
        7.3 Suboptimal Method: Heuristic Distribution
        7.4 Genetic Algorithm for Pattern Allocation
        7.5 Comparison between Genetic Algorithm and 
            Mixed Integer Programming (MIP)
        7.6 Inclusion of `A Priori' Information
        7.7 GA with the Proposed Stopping Criterion versus MIP

        Appendices

        A: Comparison Between Pipelined Ring Topology and Ring Topology
        B: A Sample Parallel C Program
        C: The Branch and Bound Method for Solving Mixed Integer 
           Programming problems

        Bibliography

        Index

Readership
----------

Graduate students, research scientists and practicing engineers in
artificial neural networks and parallel computing.

Bibliographic Data:
-------------------

Title:
        Parallel Implementations of Backpropagation Neural Networks on
        Transputers -  A Study of Training Set Parallelism


Authors:
        P Saratchandran (School of Electrical & Electronic Engg, 
                         Nanyang Technological University, Singapore)
        N Sundararajan  (School of Electrical & Electronic Engg,
                         Nanyang Technological University, Singapore)
        Shou King Foo   (School of Electrical & Electronic Engg,
                         Nanyang Technological University, Singapore)

Publisher:
        World Scientific Publishing Co.

ISBN:
        981-02-2654-3

Pages:
        250 pages (Approx)

Series Title:
        Progress in Neural Processing.

Series Advisors:
        Alan Murray       (University of Edinburgh)
        Lionel Tarassenko (University of Oxford)
        John Taylor       (King's College, London)
        Andreas Weigend   (University of Colorado)

List Price:
        US$38.00, UK Pounds 26.00

Order Information:
------------------
World-Wide Web: http://www.singnet.com.sg/~wspclibk or
                http://www.wspc.co.uk

Please send orders to your regular book supplier or directly to your
nearest World Scientific office:

USA:
        World Scientific Publishing Co., Inc.
        1060 Main Street, River Edge, NJ 07661, USA
        Fax: 1-201-487-9656 Tel: 1-201-487-9655 Toll-free: 1-800-227-7562

UK:
        World Scientific Publishing Co., Ltd.
        57 Shelton Street, Covent Garden, London WC2H 9HE, UK.
        Fax: 44-171-836-2020 Tel: 44-171-836-0888

SINGAPORE:
        World Scientific Publishing Co. Pte. Ltd.
        Farrer Road, P O Box 128, SINGAPORE 912805.
        Cable: COSPUB Tlx: RS 28561WSPC Fax: 65-382-5919
        Tel: 65-382-5663

HONG KONG:
        World Scientific Publishing (HK) Co. Ltd.
        P O Box 72482, Kowloon Central Post Office, HONG KONG.
        Fax: 852-2-771-8155 Tel: 852-2-771-8791

INDIA:
        World Scientific Publishing Co. Pte. Ltd.
        4911, 9th Floor, High Point IV, 45 Palace Road,
        Bangalore 560 001, INDIA.
        Tlx: 0845-2900 PCO IN Fax: 91-80-334-4593 Tel: 91-80-220-5972

TAIWAN:
        World Scientific Publishing Co. Pte. Ltd.
        5F-6, No. 88, Sec 3, Hsin-Sheng S Road, Taipei, Taiwan, ROC.
        Fax/Tel: 886-2-369-1366

"RUSH ORDERS"
        in USA and Canada, Call Toll-free 1-800-227-7562
        in Europe Fax: 44-171-836-2020
        in other countries Fax: 65-382-5919

Please allow 4 to 6 weeks for delivery. Please note that our
teleordering mnemonic is WSPC. Prices subject to change without prior
notice.

(Cut Here)
---------------------------------------------------------------------

Order Form:
-----------

Title:
        Parallel Implementations of Backpropagation Neural Networks on
        Transputers - A Study of Training Set Parallelism

ISBN:
        981-02-2654-3

      ____________________________________________________
     |  Number of Copies        | Amount(US$ or UK pounds)|
     |____________________________________________________|
     |                          |                         |
     |__________________________|_________________________|
 
               
Name:____________________________________________________ 

Organization:____________________________________________

Address:_________________________________________________

City:_______________________ State:______________________

Zip:________________ Country:____________________________


Methods of Payment:
-------------------

[ ]     Cheque/Bank draft enclosed for (US$/UK Pounds)_____________
        - For cheque payment in USA, please make cheque payable to
        "World Scientific Publishing Co., Inc."
        - For cheque payment from the rest of the world, please make
        cheque payable to "World Scientific Publishing Co. Pte. Ltd."

[ ]     Charge my       [ ] VISA  [ ] MC  [ ] Amex  [ ] Diners Club

        Card No:____________________________ Exp. Date:____________

        Signature:__________________________ Tel:__________________

[ ]     Bill my company/institution:_______________________________

        _____________________________(Please attach purchase order)

--
Articles to bigrigg+parallel@pitt.edu (Administrative: bigrigg@cs.pitt.edu)
Archive: http://www.hensa.ac.uk/parallel/internet/usenet/comp.parallel



