mgj@cup.portal.com (Mark Gregory Jurik) (03/01/90)
Pre-publication report on BACKPERCOLATION
While BackPropagation creates output error gradients for each cell, BackPercolation is a second pass that creates,
or assigns, true error to each cell. With this information,
each cell has a much better "understanding" of what it needs
to do. Experimental results show amazing learning behavior
during the very difficult 8-2-8 encoder problem, such as :
1. relative insensitivity to initial condintions
2. no apparent vulnerability to local minima
3. exponential error decrease almost from the very start
4. very stable with hidden layers ... the learning rate
factor can be a full 1.0 !!
If you are interested, I will mail you a copy of the experimental
results via U.S. Post. Your request *must* include a self-
addressed stamped envelope (standard business size). I'll go
broke otherwise. If you like what you see, I'll then send you
the algorithm details.
Why this approach and not via a conference? Simply put, that
route will take too long. I'm looking for co-researchers and/or
other experimenters NOW.
Send your request to Mark Jurik
PO 2379
Aptos, CA 95001 USAlinco@eng.umd.edu (Sam Lin) (03/02/90)
In article <27433@cup.portal.com> mgj@cup.portal.com (Mark Gregory Jurik) writes: >Pre-publication report on BACKPERCOLATION > >While BackPropagation creates output error gradients for each cell, BackPercolation is a second pass that creates, >or assigns, true error to each cell. With this information, [description deleted] > >If you are interested, I will mail you a copy of the experimental >results via U.S. Post. Your request *must* include a self- >addressed stamped envelope (standard business size). I'll go >broke otherwise. If you like what you see, I'll then send you >the algorithm details. > >Why this approach and not via a conference? Simply put, that >route will take too long. I'm looking for co-researchers and/or >other experimenters NOW. > >Send your request to Mark Jurik > PO 2379 > Aptos, CA 95001 USA How about posting the algorithm to the net? It would be even faster than USnail, and save everyone lotsa stamps and envelopes. It sounds very interesting.
kbesrl@ux1.cso.uiuc.edu (03/03/90)
>In article <27433@cup.portal.com> mgj@cup.portal.com (Mark Gregory Jurik) writes: >>Pre-publication report on BACKPERCOLATION >> >>While BackPropagation creates output error gradients for each cell, BackPercolation is a second pass that creates, >>or assigns, true error to each cell. With this information, >[description deleted] >> >>If you are interested, I will mail you a copy of the experimental >>results via U.S. Post. Your request *must* include a self- >>addressed stamped envelope (standard business size). I'll go >>broke otherwise. If you like what you see, I'll then send you >>the algorithm details. >> >How about posting the algorithm to the net? It would be even faster >than USnail, and save everyone lotsa stamps and envelopes. It sounds >very interesting. I second the proposal for posting it on the net!
efrethei@blackbird.afit.af.mil (Erik J. Fretheim) (03/03/90)
kbesrl@ux1.cso.uiuc.edu writes: >>In article <27433@cup.portal.com> mgj@cup.portal.com (Mark Gregory Jurik) writes: >>>Pre-publication report on BACKPERCOLATION >>> >>>While BackPropagation creates output error gradients for each cell, BackPercolation is a second pass that creates, >>>or assigns, true error to each cell. With this information, >>[description deleted] >>> >>> >>How about posting the algorithm to the net? It would be even faster >>than USnail, and save everyone lotsa stamps and envelopes. It sounds >>very interesting. >I second the proposal for posting it on the net! Hear, hear (or is that here, here - pst it here). ef -- -- Erik J Fretheim efrethei@afit.af.mil AFIT/ENA Box 4151 (ATTN: CPT FRETHEIM) (513)255-5276 AVN785-5276 WPAFB, OH 45431 USA
mgj@cup.portal.com (Mark Gregory Jurik) (03/03/90)
I have been asked to post the Backpercolation algorithm. Unfortunately, the scientific community acknowledges the author/originator of an idea through published manuscripts/journals more so than electronic bulletin boards. If this material were posted, then anyone could quickly produce and publish a report while I am still methodically experimenting. It is not helpful to publish prematurely, especially when such reports are likely to be rejected by conference mediators. For now, a one-on-one approach is better until the time comes when a sufficient number of diverse experiments show rather conclusively its value. Then it would be correct to broadcast the results.
dhw@itivax.iti.org (David H. West) (03/07/90)
In article <27482@cup.portal.com> mgj@cup.portal.com (Mark Gregory Jurik) writes: |through published manuscripts/journals more so than electronic bulletin |boards. If this material were posted, then anyone could quickly produce |and publish a report while I am still methodically experimenting. It is |not helpful to publish prematurely, especially when such reports are likely |to be rejected by conference mediators. For now, a one-on-one approach is So write it out, date it, get it notarized and stored in a bank vault. That should be traceable enough.
jerry@festus.ksux.ksu.edu (Jerry Anderson) (03/07/90)
In article <5095@itivax.iti.org> dhw@itivax.UUCP (David H. West) writes: >So write it out, date it, get it notarized and stored in a bank >vault. That should be traceable enough. Mr. Jurik took the time to post a polite reply to the requests he had gotten over the net. He most definitely has the right to publish or not publish any way he sees fit. -:{}:- Jerry J. Anderson Computing Activities Cardwell Hall jerry@ksuvm Kansas State University jerry@ksuvm.ksu.edu Manhattan, KS 66506 ...!{rutgers,texbell}!ksuvax1!ksuvm.ksu.edu!jerry ...!<your favorite UUCP/Internet gateway>!ksuvm.ksu.edu!jerry
dhw@itivax.iti.org (David H. West) (03/07/90)
In article <25F4041E.70B@deimos.cis.ksu.edu> jerry@festus.ksux.ksu.edu (Jerry Anderson) writes: |In article <5095@itivax.iti.org> dhw@itivax.UUCP (David H. West) writes: | |>So write it out, date it, get it notarized and stored in a bank |>vault. That should be traceable enough. | |Mr. Jurik took the time to post a polite reply to the requests he had gotten |over the net. He most definitely has the right to publish or not publish any |way he sees fit. Of course. I'm just addressing (somewhat tersely) his apparent belief that emailing his results to people is somehow safer than attempting conventional publication. I don't think that it is, and if he is as concerned as he appears to be to establish his priority (in the event of his being right) he should do something similar to my suggestion whether on not he decides to disclose anything to anyone, and should probably read up on the copyright laws too.