[OpendTect_Users] OpendTect Neural Networks plugin

[Paul de Groot]

Hi Roman,

Quick-UVQ, which is OpendTect's waveform segmentation approach is very
similar to Paradigm's Stratimagic that uses a Kohonen network. The only
difference is in the training algorithm that is used to find the cluster
centers. The UVQ algorithm only updates the winner while a Kohonen network
updates the winner and its nearest neighbors. Consequently a Kohonen
network automatically sorts the cluster centers (waveforms). This is
advantage when you color code the waveforms as similar waveforms will show
up with similar colors in the segmentation map. For this reason our UVQ
algorithm also sorts the waveforms but it does this as a separate step
after the training is finished. I have captured the algorithm in two
slides:
https://drive.google.com/file/d/0Bw_jc73Z9AFGeGxLazZfcjByNFk/edit?usp=sharing.

Best regards,


Paul de Groot
President

dGB Earth Sciences
Nijverheidstraat 11-2
7511 JM Enschede
The Netherlands

T: +31 53 43 15 155
E: paul.degroot at dgbes.com
www.dgbes.com & www.opendtect.org.




On Mon, Mar 31, 2014 at 1:35 PM, Roman Matloshinskiy <
rmatloshinskiy at gmail.com> wrote:

> Dear OpendTect users!
>
> Please help me with explanation of algorithm, used in NeuralNetwork plugin
> for waveform classification - are there really used classification of shape
> of the waveform (like Total algorithm in Paradigm Stratimagic) or neural
> network classification of attribute values (as a shape of waveform
> description)?
>
> Thank you very much!
>
> Best regards,
>
> Roman
>
>
>
> _______________________________________________
> Users mailing list
> Users at opendtect.org
> http://lists.opendtect.org/mailman/listinfo/users
>
> You receive this mail because you are listed on users at opendtect.org To
> unsubscribe please go to http://lists.opendtect.org/mailman/options/usersor click here :
> http://lists.opendtect.org/mailman/options/users/paul.degroot%40dgbes.com. If you encounter any problems, please contact
> support at opendtect.org .
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opendtect.org/pipermail/users/attachments/20140331/8fc9a880/attachment.html>