Smalltalk › Cincom › VisualWorks

Neuronal Network Training

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

9 messages Options

Andre Schnoor

Neuronal Network Training

Does anybody know of a Neural Network solution in Smalltalk? I'm
especially interested in the classification of objects by symbolic
attribute vectors (learning by example).

Andre

davidbuck

Re: Neuronal Network Training

Andre Schnoor wrote:

> Does anybody know of a Neural Network solution in Smalltalk? I'm
> especially interested in the classification of objects by symbolic
> attribute vectors (learning by example).
>
> Andre

I have an implementation of Hopfield nets and BAM's (Bidirectional
Associative Memory) but I suspect it's not what you're looking for.

David Buck

Charles A. Monteiro-2

Re: Neuronal Network Training

In reply to this post by Andre Schnoor

not sure if this helps but there was a book once published entitled
"Advance Smalltalk" by jonathan Plentzke if I recall the name correctly,
code for building a Neural Net was provided. Not sure how robust that was
etc.

hth,

-Charles

On Thu, 16 Mar 2006 11:09:42 -0500, Andre Schnoor <[hidden email]>
wrote:

> Does anybody know of a Neural Network solution in Smalltalk? I'm
> especially interested in the classification of objects by symbolic
> attribute vectors (learning by example).
>
> Andre

--
Charles A. Monteiro

Mark Pirogovsky-3

Re: Neuronal Network Training

In reply to this post by Andre Schnoor

Once upon a time (90- 95) there was a big Neural net application built
on the ST80 v.4.1. The application was used for the real
client(Airline revenue management system) with reasonable enough success
so it was even nominated for the "Smithsonian Award" or something
similar in 94.

I don't have any more information, but if properly motivated, I can try
to dig up some names.

-Mark

Andre Schnoor wrote:
> Does anybody know of a Neural Network solution in Smalltalk? I'm
> especially interested in the classification of objects by symbolic
> attribute vectors (learning by example).
>
> Andre
>
>
>

stéphane ducasse-2

Re: Neuronal Network Training

In reply to this post by davidbuck

Hi david

could you publish your code on one the cincom store?
I feel that it would be good to packages of interesting libraries
available (may be porting weka one would be good).

Stef

On 16 mars 06, at 17:08, David Buck wrote:

> Andre Schnoor wrote:
>
>> Does anybody know of a Neural Network solution in Smalltalk? I'm
>> especially interested in the classification of objects by symbolic
>> attribute vectors (learning by example).
>>
>> Andre
>
> I have an implementation of Hopfield nets and BAM's (Bidirectional
> Associative Memory) but I suspect it's not what you're looking for.
>
> David Buck
>
>

Andre Schnoor

Re: Neuronal Network Training

In reply to this post by davidbuck

>> Does anybody know of a Neural Network solution in Smalltalk? I'm
>> especially interested in the classification of objects by symbolic
>> attribute vectors (learning by example).
>
> I have an implementation of Hopfield nets and BAM's (Bidirectional
> Associative Memory) but I suspect it's not what you're looking for.
>
Not really. I'm looking for a simple object classification, i.e. input a
vector of attributes (somehow mapped to numeric values) and output 1 out
of n classes, trainable by examples. Possibly a similar solution like
optical character recognition.

Andre

Stew MacLean

RE: Neuronal Network Training

Charles mentioned the Advanced Smalltalk book, which I've just had a
quick look through.

It contains an example of back-propagation, (translated directly from a
poorly written C program - (his words)).

He also mentions a "very well-architected" program by Murali Krishnan of
UIUC. Apparently the source code is in the archives and includes three
types of networks: a back-propagation, aHopfield net, and a
self-organizing feature map network. (with UI's for VW). It also allows
for other types of networks because of it's generic approach.

Maybe useful?

Cheers,

Stewart

>-----Original Message-----
>From: [hidden email] [mailto:[hidden email]]
>Sent: 17 March 2006 9:56
>To: vwnc-list
>Subject: Re: Neuronal Network Training
>
>
>>> Does anybody know of a Neural Network solution in Smalltalk? I'm
>>> especially interested in the classification of objects by symbolic
>>> attribute vectors (learning by example).
>>
>> I have an implementation of Hopfield nets and BAM's (Bidirectional
>> Associative Memory) but I suspect it's not what you're looking for.
>>
>Not really. I'm looking for a simple object classification, i.e. input

a
>vector of attributes (somehow mapped to numeric values) and output 1
out
>of n classes, trainable by examples. Possibly a similar solution like
>optical character recognition.
>
>Andre
>

Hans N Beck

Re: Neuronal Network Training

Hi,

what's about to connect to SNNS

http://www-ra.informatik.uni-tuebingen.de/SNNS/ ?

it supports a wide range of networks, connecting to Smalltalk should not
that difficult, becaus it has an API, if I remember right (I've
implemented nets some years ago for SNNS). Also it runs on many systems.

If someone would take a start for it I would help :-)

Regards

Hans

-----Original Message-----
> Date: Thu, 23 Mar 2006 05:52:36 +0100
> Subject: RE: Neuronal Network Training
> From: "Stewart MacLean"
> To: ,

> Charles mentioned the Advanced Smalltalk book, which I've just had a
> quick look through.
>
> It contains an example of back-propagation, (translated directly from
> a poorly written C program - (his words)).
>
> He also mentions a "very well-architected" program by Murali Krishnan
> of UIUC. Apparently the source code is in the archives and includes
> three types of networks: a back-propagation, aHopfield net, and a
> self-organizing feature map network. (with UI's for VW). It also
> allows for other types of networks because of it's generic approach.
>
> Maybe useful?
>
> Cheers,
>
> Stewart
>
>
>
> > -----Original Message-----
> > From: [hidden email] [mailto:[hidden email]]
> > Sent: 17 March 2006 9:56
> > To: vwnc-list
> > Subject: Re: Neuronal Network Training
> >
> >
> > > > Does anybody know of a Neural Network solution in Smalltalk? I'm
> > > > especially interested in the classification of objects by
> > > > symbolic attribute vectors (learning by example).
> > > >
> > >
> > > I have an implementation of Hopfield nets and BAM's (Bidirectional
> > > Associative Memory) but I suspect it's not what you're looking
> > > for.
> > >
> > >
> > Not really. I'm looking for a simple object classification, i.e.
> > input
> a
> > vector of attributes (somehow mapped to numeric values) and output 1
> >
> out
> > of n classes, trainable by examples. Possibly a similar solution
> > like optical character recognition.
> >
> > Andre
> >
> >
>
>
>
>

Charles A. Monteiro-2

Re: Neuronal Network Training

interesting, it apparently is mostly C right? so it DLLCC-ed. There is now
of course a Java version and they undoubtedly used JNI. The only issue
though for some may be the license.

-Charles

On Thu, 23 Mar 2006 02:39:30 -0500, [hidden email]
<[hidden email]> wrote:

> Hi,
>
> what's about to connect to SNNS
>
> http://www-ra.informatik.uni-tuebingen.de/SNNS/ ?
>
> it supports a wide range of networks, connecting to Smalltalk should not
> that difficult, becaus it has an API, if I remember right (I've
> implemented nets some years ago for SNNS). Also it runs on many systems.
>
> If someone would take a start for it I would help :-)
>
> Regards
>
> Hans
>
> -----Original Message-----
>> Date: Thu, 23 Mar 2006 05:52:36 +0100
>> Subject: RE: Neuronal Network Training
>> From: "Stewart MacLean"
>> To: ,
>
>> Charles mentioned the Advanced Smalltalk book, which I've just had a
>> quick look through.
>>
>> It contains an example of back-propagation, (translated directly from
>> a poorly written C program - (his words)).
>>
>> He also mentions a "very well-architected" program by Murali Krishnan
>> of UIUC. Apparently the source code is in the archives and includes
>> three types of networks: a back-propagation, aHopfield net, and a
>> self-organizing feature map network. (with UI's for VW). It also
>> allows for other types of networks because of it's generic approach.
>>
>> Maybe useful?
>>
>> Cheers,
>>
>> Stewart
>>
>>
>>
>> > -----Original Message-----
>> > From: [hidden email] [mailto:[hidden email]]
>> > Sent: 17 March 2006 9:56
>> > To: vwnc-list
>> > Subject: Re: Neuronal Network Training
>> >
>> >
>> > > > Does anybody know of a Neural Network solution in Smalltalk? I'm
>> > > > especially interested in the classification of objects by
>> > > > symbolic attribute vectors (learning by example).
>> > > >
>> > >
>> > > I have an implementation of Hopfield nets and BAM's (Bidirectional
>> > > Associative Memory) but I suspect it's not what you're looking
>> > > for.
>> > >
>> > >
>> > Not really. I'm looking for a simple object classification, i.e.
>> > input
>> a
>> > vector of attributes (somehow mapped to numeric values) and output 1
>> >
>> out
>> > of n classes, trainable by examples. Possibly a similar solution
>> > like optical character recognition.
>> >
>> > Andre
>> >
>> >
>>
>>
>>
>>
>
>

--
Charles A. Monteiro