home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!munnari.oz.au!uniwa!bilby.cs.uwa.oz.au!wambenger!phi
- From: phi@cs.uwa.oz.au (Philip HINGSTON)
- Subject: Re: Kohonen nets with other than rectangular output arrays
- Message-ID: <phi.726368737@wambenger>
- Sender: usenet@bilby.cs.uwa.edu.au
- Nntp-Posting-Host: wambenger
- Organization: Dept. Computer Science, University of Western Australia.
- Date: Thu, 7 Jan 1993 01:05:37 GMT
- Lines: 154
-
- Before Christmas I posted a request for references on Kohonen networks
- with neighbourhood graphs other than rectangular grids. I recieved a
- number of very helpful replies as well as requests for copies of those
- replies. Many thanks to those who responded with references. Rather than
- responding individually to those who asked for copies, I include 2 replies
- which subsumed the others.
-
- -------------------------------------------------------------------------------
- Return-Path: jonabbey@cs.utexas.edu
- Return-Path: <jonabbey@cs.utexas.edu>
- Received: from deepthought.cs.utexas.edu by cs.uwa.edu.au (4.1/SMI-4.0+cs.uwa-1.0)
- id AA15362; Tue, 22 Dec 92 12:06:08 WST
- From: jonabbey@cs.utexas.edu (Jonathan David Abbey)
- Received: from earth.cs.utexas.edu by deepthought.cs.utexas.edu (5.64/1.2/relay) with SMTP
- id AA06364; Mon, 21 Dec 92 22:05:48 -0600
- Received: by earth.cs.utexas.edu (5.64/Client-v1.3)
- id AA09098; Mon, 21 Dec 92 22:05:32 -0600
- Date: Mon, 21 Dec 92 22:05:32 -0600
- Message-Id: <9212220405.AA09098@earth.cs.utexas.edu>
- To: phi@cs.uwa.edu.au
- Subject: Re: Kohonen nets with other than rectangular output arrays
- Newsgroups: comp.ai.neural-nets
- In-Reply-To: <phi.724924574@wambenger>
- Organization: CS Dept, University of Texas at Austin
- Cc:
-
- In article <phi.724924574@wambenger> you write:
- >Hi, I guess the subject line says it. I have found some things
- >discussing the choice between 1-d, 2-d, 3-d etc arrays of output
- >neurons. What about other neighbourhood graphs? e.g. toroidal, trees etc
- >Does anyone know of any references? Thanks, phi
- >
-
- Hi, the Fritzke paper discusses a variant of the kohonen map that adjusts
- better to non-convex regions. It's got a nice reference list. The
- Blackmore and Miikkulainen paper presents a different method for the same
- purpose. Contact risto@cs.utexas.edu to inquire, if the Fritzke paper
- suits your interest.
-
- Best of luck!
-
- @string{ICANN-91 = "Proceedings of the International Conference on
- Artificial Neural Networks {\rm (Espoo, Finland)}"}
- @string{NORTH = "North-Holland"}
- @string{NORTH-ADDR = "Amsterdam; New York"}
-
- @InProceedings{fritzke:grow,
- author = "Bernd Fritzke",
- title = "Let it Grow---{S}elf-Organizing Feature Maps with
- Problem Dependent Cell Structure",
- booktitle = ICANN-91,
- pages = "403--408",
- publisher = NORTH,
- address = NORTH-ADDR,
- year = "1991"
- }
-
- @Unpublished{blackmore:incremental,
- author = "Justine Blackmore and Risto Miikkulainen",
- title = "Incremental Grid Growing: {E}ncoding High-Dimensional
- Structure into a Two-Dimensional Feature Map",
- note = "submitted to ICNN-93",
- year = "1992"
- }
- --
- Jonathan Abbey jonabbey@cs.utexas.edu
- the university of texas at austin --------------------
- computer science/math/psychology -- Oh, my.. life is just so.. you know.. --
- -------------------------------------------------------------------------------
- Return-Path: karit@idiap.ch
- Return-Path: <karit@idiap.ch>
- Received: from chx400.switch.ch by cs.uwa.edu.au (4.1/SMI-4.0+cs.uwa-1.0)
- id AA20357; Tue, 22 Dec 92 16:36:34 WST
- Received: from maya.idiap.ch by chx400.switch.ch with SMTP (PP)
- id <22524-0@chx400.switch.ch>; Tue, 22 Dec 1992 09:34:52 +0100
- Received: from dolent.idiap.ch by idiap.ch (4.1/SMI-4.1) id AA06883;
- Tue, 22 Dec 92 09:34:49 +0100
- Date: Tue, 22 Dec 92 09:34:49 +0100
- From: karit@idiap.ch (Kari Torkkola)
- Message-Id: <9212220834.AA06883@idiap.ch>
- Received: by dolent.idiap.ch (4.1/SMI-4.1) id AA08681;
- Tue, 22 Dec 92 09:34:46 +0100
- To: phi@cs.uwa.edu.au (Philip HINGSTON)
- Subject: Re: Kohonen nets with other than rectangular output a
- Cc: karit@idiap.ch
- Reply-To: karit@idiap.ch
- Status: RO
-
- In article <phi.724924574@wambenger>, you write:
- |> Hi, I guess the subject line says it. I have found some things
- |> discussing the choice between 1-d, 2-d, 3-d etc arrays of output
- |> neurons. What about other neighbourhood graphs? e.g. toroidal, trees etc
- |> Does anyone know of any references? Thanks, phi
- |>
-
-
-
- You might want to look at the following articles:
-
-
- @Article{Kangas90,
- author = "Jari Kangas and Teuvo Kohonen and Jorma Laaksonen",
- title = "Variants of Self-Organizing Maps",
- journal = "IEEE Transactions on Neural Networks",
- year = "1990",
- volume = "1",
- number = "1",
- pages = "93-99",
- month = "March",
- }
-
- @InProceedings{Martinetz91,
- author = "Thomas Martinetz and Klaus Schulten",
- title = "A Neural-Gas Network Learns Topologies",
- booktitle = "Artificial Neural Networks
- (Proceedings of the Int. Conf. on
- Artificial Neural Networks, ICANN91)",
- year = "1991",
- volume = "I",
- pages = "397-402",
- address = "Espoo, Finland",
- month = "June 24-28",
- }
-
- @InProceedings{Fritzke91,
- author = "Bernd Fritzke",
- title = "Let It Grow - Self-Organizing feature Maps with
- Problem Dependent cell Structure",
- booktitle = "Artificial Neural Networks
- (Proceedings of the Int. Conf. on
- Artificial Neural Networks, ICANN91)",
- year = "1991",
- volume = "I",
- pages = "403-408",
- address = "Espoo, Finland",
- month = "June 24-28",
- }
-
-
- Once I was using SOMs as the basis for speech recognition. Then
- I played myself with toroids and spherical surfaces, but they
- affected the recognition accuracy (which was all I was interested in
- at that point) very little, if at all (as expected).
-
-
- Yours,
-
- Kari Torkkola, research director, speech recognition
- IDIAP (Institut Dalle Molle
- d'Intelligence Artificielle Perceptive)
- C.P. 609 tel: +41-26-227664
- CH-1920 Martigny fax: +41-26-227818
- Switzerland email: karit@idiap.ch
- -------------------------------------------------------------------------------
-