Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!news.duke.edu!eff!usenet.ins.cwru.edu!howland.reston.ans.net!EU.net!uunet!psinntp!hk.super.net!wanchai.hk.super.net!news.ust.hk!hpg30a.csc.cuhk.hk!eng_ser1!lwchan
From: lwchan@cs.cuhk.hk (CHAN Lai-Wan)
Subject: Re: Guardline to choose learning rate
Message-ID: <Cy9nt9.AF5@eng_ser1.erg.cuhk.hk>
Lines:  26
Sender: news@eng_ser1.erg.cuhk.hk
Organization: Engineering Faculty, The Chinese U. of Hong Kong
References: <1994Oct25.050559.1720@uxmail.ust.hk>
Date: Wed, 26 Oct 1994 06:10:20 GMT

cs_leo@ug.cs.ust.hk (Wong Tsz Cheong) writes:

>Hi all,
>	I am a final year ug student and doing a project about the backprogation Neural
>	Network. Could some one tell me how I can find the guardline of choosing learning
>	rate and momentum values? Whatever book or paper are welcome?
>	Thank for your help!
>Leo.

I have a paper on how to automatically change the learning rate and
momentum term during the learning, i.e. users don't need to decide the
optimum learning rate.  It is in :

An Adaptive Training Algorithm for Back Propagating Networks, L-W.
Chan & F. Fallside, Computer Speech and Language, 2, 3, Academic
Press, 1987, p205-218.


laiwan

--
===================================================================
Laiwan Chan, Computer Science Dept, The Chinese Univ. of Hong Kong.
~{3B@vTF~} ~{Oc8[VPND4sQ'5gWS<FKcQ'O5~}          Shatin, Hong Kong.	
Tel: 609 8434  Fax: 603 5024  email: lwchan@cs.cuhk.hk 	      
