flat assembler
Message board for the users of flat assembler.

Index > Projects and Ideas > [IDEA] Neural Networks simulator - project is sleeping

Author
Thread Post new topic Reply to topic
Madis731



Joined: 25 Sep 2003
Posts: 2140
Location: Estonia
Madis731
Hi,
This project allows you to define Neural Networks in the source file and compile it. The definitions are slightly documented - you can get hints from the other samples. The basic idea is to define a neuron with a number of inputs where each input has a weight. The program only needs this data to make a neural network and calculate the outcome.

This isn't even alpha release so don't expect too much from it. This converts a simple XOR into NN. The float format is improved to accept:
Code:
-14,25
00240000,0003053000
24.53
-.5
.00243520
5
-6
    

..but don't try..
Code:
6.
-3.
4..5--.
;etc Razz
    

..because even I get confused, what the user needs in these cases Smile

The only numbers in this examples you would use are -1,0 == ZERO and 1,0 == ONE and the result can be read from the OUT1: 1 or 3 meaning ZERO or ONE

The OUT2 is a debug, it just shows the value in the middle layer (see the picture).

STEP-button only updates all neurons ONCE!!! Propagating may take time so press 2 times to get the input value to the ouput. This is a rare case with two layers and my numbering is from top down so the value gets propagated to the last layer in one pass.

EDIT:
btw, if anyone has ideas on the following subjects:
FIXED:1) Reading loose format user input floating-point numbers
±01230,01230
±01230.01230
±,01230
±.01230
with stripping leading/trailing zeros support.

2) Another format to save NNs
2.1) Current implementation:
pointerArray rd numberOfNeurons
Neuron dd connectionCount,currentValue, Count*<inputFromNeuron,weight>

2.2) Another possibility is to make an array of connections (src,dst)
and their weights. Then the length of each neuron and every connection
is unified and can be easily read. On the downside, the lookup of neurons
and their values, weights, etc. is algorithmically O(n²) so using such
a nework in real-time application is not applicable Sad


Description: XOR NN schema, a copy from a PDF about NNs. I'll do it myself if anyone starts to complain about copyright :P
Filesize: 2.32 KB
Viewed: 7743 Time(s)

XOR.png


Description: Neural Networks mimicking XOR function
Download
Filename: NN.7z
Filesize: 3.25 KB
Downloaded: 640 Time(s)


_________________
My updated idol Very Happy http://www.agner.org/optimize/


Last edited by Madis731 on 08 Sep 2006, 10:04; edited 3 times in total
Post 29 Jan 2006, 16:01
View user's profile Send private message Visit poster's website Yahoo Messenger MSN Messenger Reply with quote
vid
Verbosity in development


Joined: 05 Sep 2003
Posts: 7105
Location: Slovakia
vid
Madis: how is this thing looking? Do you think it should remain here in project section?

some neural networks discussed here
Post 08 Sep 2006, 07:08
View user's profile Send private message Visit poster's website AIM Address MSN Messenger ICQ Number Reply with quote
Madis731



Joined: 25 Sep 2003
Posts: 2140
Location: Estonia
Madis731
Well, I've developed it a bit, but I haven't uploaded a newer version yet. Where would you move it?

EDIT: The topic slightly modified and few lines added to describe it. Project itself is hibernated, but will be blown life into in the near future (months)


Last edited by Madis731 on 08 Sep 2006, 07:46; edited 2 times in total
Post 08 Sep 2006, 07:34
View user's profile Send private message Visit poster's website Yahoo Messenger MSN Messenger Reply with quote
vid
Verbosity in development


Joined: 05 Sep 2003
Posts: 7105
Location: Slovakia
vid
okay, so write some brief description WHAT this is at beginning. is it some engine simulating NN, or what?
Post 08 Sep 2006, 07:37
View user's profile Send private message Visit poster's website AIM Address MSN Messenger ICQ Number Reply with quote
masonswanson



Joined: 17 Aug 2010
Posts: 51
masonswanson
Is This Dead? I Would Really Like To Collaborate On A Project Like This
Post 26 Dec 2010, 07:43
View user's profile Send private message Send e-mail Visit poster's website Yahoo Messenger Reply with quote
Madis731



Joined: 25 Sep 2003
Posts: 2140
Location: Estonia
Madis731
Yes, I'm afraid so, I haven't had the time to continue this. I guess one needs to use many of those open-source-java-based NN simulators and calculators. The idea was bigger than me Smile
I don't want to write 'done' or anything like it there. So it just sits there
Post 26 Dec 2010, 08:39
View user's profile Send private message Visit poster's website Yahoo Messenger MSN Messenger Reply with quote
Display posts from previous:
Post new topic Reply to topic

Jump to:  


< Last Thread | Next Thread >
Forum Rules:
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You cannot attach files in this forum
You can download files in this forum


Copyright © 1999-2020, Tomasz Grysztar. Also on GitHub, YouTube, Twitter.

Website powered by rwasa.