I am currently developing a neural networks project in D.
Now it only supports feed-forward networks and has the following features
implemented:

    -TrainingAlgorithm
        -Backpropagation Classic
        -Backpropagation Adaptive (Silva & Almeida)
        -Backpropagation DeltaBarDelta
        -Backpropagation QuickProp
        -Backpropagation Rprop
        -Backpropagation SuperSab

    -ActivationFunction
        -Binary
        -Gaussian
        -Linear
        -SemiLinear
        -Tanh

    -CostFunction
        -Absloute
        -CrossEntropy
        -Quadratic
        -DualSlope

    -CostControl (backtrack on failure)

    -Regularization (weight decay)
        -Absolute
        -Quadratic

    -Normalization
        -Decimal
        -Logarithmic
        -MinMax
        -Sigmoidal
        -SoftMax
        -Zscore

I intend to implement in a near future:
    -Early stop (split-sample & cross-validation)
    -Add input noise (regularization method)
    -Load/save methods to make the network persistent
    -Gui (if i have knowledge to do it)

I wanna know if the D community is interested in this area, in particular in
this project.
If so, I could put the source available.

-- 
Miguel Ferreira Simoes

Greetings Miguel,

I'm really interested in and very happy if you can provide the source.
Unfortunately I'm not in the position to invest much time in D and the project
:((   , but this will change next year (hopefully).

If your planning to set your Project under the GPL, maybe you can put it to dsource.org ... or you'll send the source to bjoern +dot+ diwald /at/ gmx *dot* de

Bye
Björn


In article <cna375$22si$1@digitaldaemon.com>, Miguel Ferreira Simões says...
>
>I am currently developing a neural networks project in D.
>Now it only supports feed-forward networks and has the following features
>implemented:
>
>    -TrainingAlgorithm
>        -Backpropagation Classic
>        -Backpropagation Adaptive (Silva & Almeida)
>        -Backpropagation DeltaBarDelta
>        -Backpropagation QuickProp
>        -Backpropagation Rprop
>        -Backpropagation SuperSab
>
>    -ActivationFunction
>        -Binary
>        -Gaussian
>        -Linear
>        -SemiLinear
>        -Tanh
>
>    -CostFunction
>        -Absloute
>        -CrossEntropy
>        -Quadratic
>        -DualSlope
>
>    -CostControl (backtrack on failure)
>
>    -Regularization (weight decay)
>        -Absolute
>        -Quadratic
>
>    -Normalization
>        -Decimal
>        -Logarithmic
>        -MinMax
>        -Sigmoidal
>        -SoftMax
>        -Zscore
>
>I intend to implement in a near future:
>    -Early stop (split-sample & cross-validation)
>    -Add input noise (regularization method)
>    -Load/save methods to make the network persistent
>    -Gui (if i have knowledge to do it)
>
>I wanna know if the D community is interested in this area, in particular in
>this project.
>If so, I could put the source available.
>
>-- 
>Miguel Ferreira Simoes
>
>

Very interested!

I would like to help as well ( was doing this over the weekend :) ) , maybe we
can get a whole AI suite going.

Charlie

In article <cna375$22si$1@digitaldaemon.com>, Miguel Ferreira Simões says...
>
>I am currently developing a neural networks project in D.
>Now it only supports feed-forward networks and has the following features
>implemented:
>
>    -TrainingAlgorithm
>        -Backpropagation Classic
>        -Backpropagation Adaptive (Silva & Almeida)
>        -Backpropagation DeltaBarDelta
>        -Backpropagation QuickProp
>        -Backpropagation Rprop
>        -Backpropagation SuperSab
>
>    -ActivationFunction
>        -Binary
>        -Gaussian
>        -Linear
>        -SemiLinear
>        -Tanh
>
>    -CostFunction
>        -Absloute
>        -CrossEntropy
>        -Quadratic
>        -DualSlope
>
>    -CostControl (backtrack on failure)
>
>    -Regularization (weight decay)
>        -Absolute
>        -Quadratic
>
>    -Normalization
>        -Decimal
>        -Logarithmic
>        -MinMax
>        -Sigmoidal
>        -SoftMax
>        -Zscore
>
>I intend to implement in a near future:
>    -Early stop (split-sample & cross-validation)
>    -Add input noise (regularization method)
>    -Load/save methods to make the network persistent
>    -Gui (if i have knowledge to do it)
>
>I wanna know if the D community is interested in this area, in particular in
>this project.
>If so, I could put the source available.
>
>-- 
>Miguel Ferreira Simoes
>
>

Charlie wrote:
> Very interested!
> 
> I would like to help as well ( was doing this over the weekend :) ) , maybe we
> can get a whole AI suite going.
> 
> Charlie
> 
> In article <cna375$22si$1@digitaldaemon.com>, Miguel Ferreira Simões says...
> 
>>I am currently developing a neural networks project in D.
>>Now it only supports feed-forward networks and has the following features implemented:
>>
>>   -TrainingAlgorithm
[snip an impressive list of features]
>>I intend to implement in a near future:
>>   -Early stop (split-sample & cross-validation)
>>   -Add input noise (regularization method)
>>   -Load/save methods to make the network persistent
>>   -Gui (if i have knowledge to do it)
>>
>>I wanna know if the D community is interested in this area, in particular in this project.
>>If so, I could put the source available.

Hi Miguel! I would be very interested, as I have been twiddling a little with the idea of making a neuro-fuzzy toolbox a la Matlab's ANFISEDIT in Java for teaching purposes (only better ;-) ). Fortunately I have only been thinking about the rough design so far, so maybe it's time to switch to a more efficient language.

Just like Björn (in another post) I don't have much time to spare on development right now, but that will hopefully change at some point over the next few months.

BTW, I'd like to take this opportunity to thank you for sharing the XML module code about a month ago - I downloaded it from the newsgroup and looked at it, but haven't had much chance to play around with it yet. It looks promising, though.

Cheers,
Jan

Ok. I am glad to see interested people on the project! I will spent this week and weekend implementing new features and then I will put the source code available.

Probably the dsource.org is the best place isn't it?

Miguel Ferreira Simoes