Xref: utzoo comp.ai.neural-nets:621 sci.philosophy.tech:1111
Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!purdue!gatech!myke
From: myke@gatech.edu (Myke Rynolds)
Newsgroups: comp.ai.neural-nets,sci.philosophy.tech
Subject: Re: request for philosophic reactions to connectionism
Keywords: connectionism philosophy representations
Message-ID: <18503@gatech.edu>
Date: 22 Apr 89 00:58:19 GMT
References: <370@eurtrx.UUCP| <18496@gatech.edu| <1738@Portia.Stanford.EDU|
Reply-To: myke@gatech.UUCP (Myke Rynolds)
Organization: School of Information and Computer Science, Georgia Tech, Atlanta
Lines: 52

At portia?! Hey, do you know Paul Gunnels?
Chris Kortge writes:
|Myke Rynolds writes:
||I think that BAMs (bi-direction associative memories) and it's conceptual
||parent, ART (adaptive resonance theory) give a profound critique of the
||connectionist models. Grossberg, the inventer of ART way back in '76, goes
||into great detail about how nothing anyone in the connectist school of thought
||has said is new, or even as powerful as what already exists! ART is proven
||to converge on any complexity of input, no connectionist model can claim this.
||They can learn only by limiting the complexity of the input, thus the failure
||of bp to deal with large and complex systems.
||
|Hold on a second!  Why is it, then, that people are using
|back-propagation learning on most practical applications?
Good question. Maybe its fad?

|I agree that
|bp has trouble with large systems, but it's important to look at the
|*results* of the learning process, too.  BP can learn distributed
|representations, which have advantages over strictly categorical ones,
|which is what ART learns.
False! ART learns internal reps. Both BP and ART generate their own internal
reps (for no good reason in my opinion), but BAMs simply associate input
vectors with output vectors.

|More importantly, since BP does supervised
|learning, its internal representation is automatically suited to the
|task at hand; ART is unsupervised, and thus it's categories are not
|necessarily useful for facilitating the required outputs.
But unless the superviser is omniscient, it doesn't know when to stop being
plastic to prevent memory washout. ART does not suffer from this. The lack
of need for a superviser is not a weakness, it is a tremendous advantage!
|
||For all its greater power, [ART] is much much simpliar than these other models
||that cloud the issue with ad hoc hockus pockus. Grossberg's model is nothing
||more than matrix multiplication. [...]
|
|Then why can't I understand his papers?  (Don't answer that :-))
|Most likely, it's because I'm a connectionist, and
Cuz the man is lost in his own little world. However, hes not being swept along
by any mobs either.
|
||Connectionists are generally psychologists and computer scientists who do not
||appreciate the deeper simplicity of math under the outer tremendous diversity.
|
|Well, be patient with us, okay?
Ok, as long as y'all see the light soon!
-- 
Myke Rynolds
School of Information & Computer Science, Georgia Tech, Atlanta GA 30332
uucp:	...!{decvax,hplabs,ncar,purdue,rutgers}!gatech!myke
Internet:	myke@gatech.edu