Neural nets creating neural nets

Disclaimer: I know very little about neural nets.

That said, from what I understand, the optimal parameters for a neural
net vary from task to task: the number of nodes, layers, type of
feedback, etc.  I gather that it is largely up to the designer's sense
of what would be appropriate for any given problem, and then some trial
and error.

Has anyone fooled around with having a neural net optimize a neural
net?  It would give a configuration as an output, the suggested net
would then run on the original problem for a while, and its trained
accuracy would be given as input to the meta-net.  Is such an
arrangement feasible?

0
cappallo (3)
12/7/2006 9:25:55 PM
comp.ai.neural-nets 5682 articles. 1 followers. tomhoo (63) is leader. Post Follow

2 Replies
281 Views

Similar Articles

[PageSpeed] 36

On 7 Dec 2006 13:25:55 -0800, "TC" <cappallo@gmail.com> wrote:

>Has anyone fooled around with having a neural net optimize a neural
>net?  It would give a configuration as an output, the suggested net
>would then run on the original problem for a while, and its trained
>accuracy would be given as input to the meta-net.  Is such an
>arrangement feasible?

My neural network software has a nested neural network to work out the
starting points for the number of hidden nodes and controls. Beyond
that point optimisation is by trial and error. For most training sets
the starting point for hidden layers should be no nodes at all. The
nested neural network never produces a zero nodes starting point and
so it can be manually overridden.

-- 
Steve Wolstenholme     Neural Planner Software

EasyNN-plus. The easy way to build neural networks.
http://www.easynn.com
0
steve
12/7/2006 10:37:25 PM
"TC" <cappallo@gmail.com> wrote in news:1165526754.966717.270370
@n67g2000cwd.googlegroups.com:

> Disclaimer: I know very little about neural nets.
> 
> That said, from what I understand, the optimal parameters for a neural
> net vary from task to task: the number of nodes, layers, type of
> feedback, etc.  I gather that it is largely up to the designer's sense
> of what would be appropriate for any given problem, and then some trial
> and error.
> 
> Has anyone fooled around with having a neural net optimize a neural
> net?  It would give a configuration as an output, the suggested net
> would then run on the original problem for a while, and its trained
> accuracy would be given as input to the meta-net.  Is such an
> arrangement feasible?
> 


If you can formulate the problem in the form of input-output (desired) datasets, sure it can be done. 
The problem is that such an approach is very problem-specific, since you would have to train a large 
set of NN configurations ("input") and relate them with their performance after normal training 
("output"), so that a meta-NN could be trainned to predict the "probably best" configuration, but 
ALWAYS for the same problem. If the problem changes, NN configurations and performances 
become irrelevant.

Perhaps you'd want to take a closer look in other similar approaches for automated NN optimization, 
such as genetic or evolutionary algorithms. Usually they are very slow in practice, but they are 
problem-independent in terms of performance.


-- 
Harris
0
Harris
12/10/2006 12:39:31 AM
Reply:

Similar Artilces:

integrate fuzzy with neural net (fuzzy in neural net)
I'm electrical engineering student..n i have to do fuzzy neural network for my final project..but,,i'm absolutely have no idea about it... so, is there anyone who can help me to shOw how to integrate fuzzy in neural network using matlab fuzzy n neural toolbox... oya,,i want to figure out that i use it for classifying needs....... thanks... ;) ...

Where one neural net learns the weights for another neural net
You can try out the idea if you like, it's here: http://code.google.com/p/lemontree/downloads/list ...

Where one neural net learns the weights for another neural net
I did think up this new system for OCT based neural nets where one neural net learns the weights for another neural net. This may provide an alternative to back propagation for such nets. Back propagation does not work too well with them. Anyway it is for you to evaluate the system I present. http://code.google.com/p/lemontree/downloads/list Sean O'Connor On 9-May-2009, Lemon Tree <bitterlemon40@yahoo.ie> wrote: > I did think up this new system for OCT based neural nets where one > neural net learns the weights for another neural net. Where does the other network get i...

Why are Neural Nets not AI?
Hi there, I made an appointment to see one of my lecturers a few days back and said it was going to be about AI. When I got there, I asked him about Neural Nets and he told me that NNs are not considered a branch of AI. I didnt want to disagree but he did not explain why. Could someone please elaborate and explain why NNs are not considered a bramch of AI? Thanks Allan "Allan Bruce" <abruce@TAKEAWAYcsd.abdn.ac.uk> wrote in message news:bg5k4f$kr5$1@news.abdn.ac.uk... > Hi there, > > I made an appointment to see one of my lecturers a few days back and said it &g...

a very simple model of neural net in neural toolbox
hey people, i'm very new to 'neural toolbox' in matlab. i must learn something. i've got some data from 2 sets of experiments with some bio-mass tank. the task is to put datas into "pre" neural network to see if it can train or create something. i've got INPUTS of tank: -one vector of 40 elements [a b c ...] (it's volume of air needed in combustion process) -one vector of 1 element [x] (it's a weight in kilos of fuel=biomass) 'ive got OUTPUTS of tank: -two not-similar vectors of 40 elements [A B C...] (one is temperature o...

a very simple model of neural net in neural toolbox
hey people, i'm very new to 'neural toolbox' in matlab. i must learn something. i've got some data from 2 sets of experiments with some bio-mass tank. the task is to put datas into "pre" neural network to see if it can train or create something. i've got INPUTS of tank: -one vector of 40 elements [a b c ...] (it's volume of air needed in combustion process) -one vector of 1 element [x] (it's a weight in kilos of fuel=biomass) 'ive got OUTPUTS of tank: -two not-similar vectors of 40 elements [A B C...] (one is temperature o...

Neural nets
I havent found a lot of information on current happenings relating to implementation of neural networks on FPGA. What I am really looking for right now is a real world application which would actually exploit the reconfigurability aspect of the FPGA. One of those many fields holding interest were Density enhancement using FPGA. This is much based on work published by Bringham Young university, but since the hardware available would allow us to work a lot more in depth, and the reconfiguration time has gone down a lot, I am looking to see if the density of neurons can further be enhanced by p...

neural net
I am trying to create a network. but i do not know how to set input waits and layer waits ! i tried by directly setting the IW and LW matrices but nothing worked for when i type disp(net) i cannot see the new weights is there a way to set the new waits On Aug 29, 2:04=A0pm, "lakshmi yeswanth" <lakshmiyeswa...@gmail.com> wrote: > I am trying to create a network. but i do not know how to set =A0input wa= its and layer waits ! > i tried by directly setting the IW and LW matrices but nothing worked for= when i type > disp(net) i cannot see the new weights is ...

Neural Net
1 name: Derek Smart He he, does anyone remember his claims? He is insisting it is true. But can somebody explain to me shortly, what was the basis of his claims? Is is theoretically possible to incorporate NN into a game like this ? Maybe he thought that a game-engine IS a neural net. And i think he watched too much Matrix - i remember when he talked that the "wood,rocks...etc." have AI. Like in the movie,where the "Oracle" said that there are programs designed for the Sunset/sundown , wind and trees ... ;) ;) Anyway, any thoughts? Cheers, T. ...

.net or not .net
Hi all I not read the group for a week ot two and i get back and suddenly there are lots of .net related messages posted. I have questions...... 1. Can someone please explain to me exactly what .net consists of and how it effects application developement. talk of .net has been around for a long time, firstlt linked with microsoft wanting application users to rent software over the internet and pay per use, but that was a long time ago. 2. Where can i learn about .net, i haven't got my head aroun COM and interaces, etc yet! Thanks all Vertuas Vertuas...

Why are Neural Nets not AI? #2
Hi there, I made an appointment to see one of my lecturers a few days back and said it was going to be about AI. When I got there, I aske dhim about Neural Nets and he told me that NNs are not considered a branch of AI. I didnt want to disagree but he did not explain why. Could someone please elaborate and explain why NNs are not considered a bramch of AI? Thanks Allan [ comp.ai is moderated. To submit, just post and be patient, or if ] [ that fails mail your article to <comp-ai@moderators.isc.org>, and ] [ ask your news administrator to fix the problems with your system. ] Allan ...

How to apply a neural net?
Hi, I'm still trying to find my way around neural nets and have managed to create a training set and trained the network, but how do you test/apply it? I'm using JavaNNS and can save the network to a file, but then how do I use the network weightings on new data? Any input appreciated, Chris. On Thu, 27 Jul 2006 10:06:40 +0100, Chris <ithinkiam@gmail.com> wrote: >Hi, > >I'm still trying to find my way around neural nets and have managed to >create a training set and trained the network, but how do you >test/apply it? > Use examples that were not inclu...

neural net question
I am using the nftool to predict temperature from serial day number. It seems to be working ok. I would like to set limits for the simulated data. Is that possible? I mean when I use the sim function to predict temperatures I want it to be limited by [0 28] or something similar. Thanks On Apr 8, 4:17=A0pm, "David Kowalske" <dkowal...@umassd.edu> wrote: > I am using the nftool to predict temperature from serial day number. =A0I= t seems to be working ok. =A0I would like to set limits for the simulated d= ata. =A0Is that possible? =A0I mean when I use the sim function to ...

neural net result
good morning everybody i have a small query i have designed a time delayed back propagation for my set of data, i have 10 set of data (basically time series),; now the problem is whenever i train the network with a data immideately it gives a good output , but after training with a new data the output in no more good. the network only works good with the last trained data set, i have 500 sample in each of the data set and i have used 25 neuron can someboby help me ...

Web resources about - Neural nets creating neural nets - comp.ai.neural-nets

Londonistan: How Britain is Creating a Terror State Within - Wikipedia, the free encyclopedia
The book encompasses a critique of multiculturalism , alleged weak policing, cultural relativism , and what Phillips calls a 'victim culture'. ...

Facebook’s Matt Kelly Describes Process Of Creating Plugin For WordPress
The plugin for WordPress launched by Facebook in June has been downloaded more than 180,000 times, and Engineer, Developer Relations Matt Kelly ...

Creating Online Ads that Drive Brand Awareness
New research shows how the creative elements of online ads impact brand awareness.

Bug - Creating a Video.Watches Action stopped working - Facebook-Entwickler
Since today executing a POST on a me/video.wacthes endpoints no longer works, it was working around 10 hours ago since then the call always returns ...

Facebook Creating Fourth-Party Privacy Rules for Facebook Connect Widget & Plugin Developers
In a recently-written post on the Facebook Developers Wiki, Facebook had added new rules for information sharing between websites that have implemented ...

The Obama campaign won by creating their own Facebook, with unique features for political campaigns.
... seen as too dangerous to nominate someone who is so thoroughly narcissistic, arrogant and dishonest. The Obama campaign won basically by creating ...

Creating A Great Startup Culture
... with any Bay Area startup these days (ping pong, beer on tap, catered lunches and Macbook Airs) – though these things are, of course, Creating ...

WIRED Creating the Nebula for iPad on the iTunes App Store
Get WIRED Creating the Nebula on the App Store. See screenshots and ratings, and read customer reviews.

Nets of freedom creating mesh networks 03.07.2013 - Flickr - Photo Sharing!
Explore Strelka Institute photo's photos on Flickr. Strelka Institute photo has uploaded 14526 photos to Flickr.

Creating the incorrigible Cartman - YouTube
"South Park" creators Trey Parker and Matt Stone were big fans of "All in the Family," a 1970s CBS sitcom. Perhaps this is why their most incorrigible ...

Resources last updated: 2/7/2016 4:58:15 PM