home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!stanford.edu!lll-winken!elroy.jpl.nasa.gov!sdd.hp.com!wupost!darwin.sura.net!paladin.american.edu!auvm!BEN.DCIEM.DND.CA!MMT
- From: mmt@BEN.DCIEM.DND.CA
- Newsgroups: bit.listserv.csg-l
- Subject: Re: Misc catching up
- Message-ID: <9208200019.AA21021@chroma.dciem.dnd.ca>
- Date: 20 Aug 92 00:19:28 GMT
- Sender: "Control Systems Group Network (CSGnet)" <CSG-L@UIUCVMD.BITNET>
- Lines: 28
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
-
- [Martin Taylor 920819 2010]
- (Avery Andrews 920820.0945) A quick response, indeed!
-
- I imagine Rick will respond to this, but in case not:
-
- > At a very minimum, PCT depends on there being consistent
- >correlations between the direction of gross effort and the direction of
- >change of net result, & a continuous function relating the magnitudes of
- >these two quantities.
-
- Only at a very high level, if there. A hierarchic control structure can
- accommodate a world the reverses the direction of effect of actions. A
- reorganizing system can learn to do it more efficiently if the reversal is
- maintained, but a fixed system can be developed that can invert the
- sign of its output if the error is increasing. Bill posted a note a
- little while ago showing how a 2-way ECS made from a pair of one-way ECSs
- could be provided with a signal that affected its gain. The same principle
- could be used in building a reverser control, I think. Anyway, humans have
- no problem with such reversals, as Rick (with Bill? I forget) showed.
-
- I think that consistency is probably required over a time commensurate with
- the inverse bandwidth of the feedback loop concerned, but that's only
- hazarding a guess. There is no requirement for a continuous function,
- either. But I think there probably is a requirement for some measure of
- closeness between percept and reference.
-
- Analysis is simpler when the relations are continuous and consistent, and
- we usually talk about that kind of system.
-