[Insight-users] NeuralNet Questions

[Raghu Venkatram]

Hi Stephanie



On 12/9/05, Stephanie Powell <stephanie-powell-1 at uiowa.edu > wrote:
>
> Hi,
>
>   I have implemented the OneHiddenLayerBackPropagationNeuralNetwork, but I
> have several questions about specific variables used in the NeuralNet
> class.
>
>  Several of the variables & the use of these variables is unclear to me
> when using a three layer neural net.
>



  In the itkOneHiddenLayer... the user has the ability to set the hidden and
> output layer bias (setting the bias) of the weight layers, but it appears as
> though this variable is not used at all.  There is a variable called m_Bias
> that belongs to the BackPropagationLayer class which is used (it's default
> value is 1), but the value itself is never set.  Why is the weights bias
> value adjusted when it's not used and the layer bias value not adjusted when
> it is used in training?
>

The weight corresponding to the bias value is adjusted. the bias itself is
not adjusted, so when you forward propagate, it is a constant value of
1.0or whatever you set it to using SetBias. But, yes, the bias in the
backpropagation layer should be set in the onehiddenlayernetwork, instead of
setting it in the weightset.

>
>

  Also, when training with momentum, it appears as though the momentum is
> defaulted to 0.9, but there is no functionality allowing the user to
> change this value.  Was it assumed that the user would not want to adjust
> this value as a training parameter?
>


thank you for pointing this out, I will commit the necessary fixes soon.
The network IO is almost complete. I will email you the files ASAP.

Your input is really valuable.

As for the variables, we will soon have a chapter in the ITK Software guide,
explaining all the variables in detail.

Also please let me know if you would like any other functionality.

Thanks,
Raghu



On 12/9/05, Stephanie Powell <stephanie-powell-1 at uiowa.edu> wrote:
>
> Hi,
>
>   I have implemented the OneHiddenLayerBackPropagationNeuralNetwork, but I
> have several questions about specific variables used in the NeuralNet
> class.
>
>   Several of the variables & the use of these variables is unclear to me
> when using a three layer neural net.
>
>   In the itkOneHiddenLayer... the user has the ability to set the hidden
> and output layer bias (setting the bias) of the weight layers, but it
> appears as though this variable is not used at all.  There is a variable
> called m_Bias that belongs to the BackPropagationLayer class which is used
> (it's default value is 1), but the value itself is never set.  Why is the
> weights bias value adjusted when it's not used and the layer bias value not
> adjusted when it is used in training?
>
>   Also, when training with momentum, it appears as though the momentum is
> defaulted to 0.9, but there is no functionality allowing the user to
> change this value.  Was it assumed that the user would not want to adjust
> this value as a training parameter?
>
> Thanks,
>   Stephanie
>
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20051210/1061cf06/attachment.htm