Re: [Discuss-gnuradio] [SOCIS] TensorFlow

[Christopher Richardson]

Hi Stefan,

Thanks for your reply.

I'm currently working on a GNU Radio block which will be able to load any TensorFlow

graph, which would allow different types of models to be utilised.

At the moment in my repository only a single hidden layer is utilised with no convolutional

layers.

cheers

Chris

On Wed, Jun 29, 2016 at 1:23 PM, Stefan Wunsch <address@hidden> wrote:

Hi,

The classification accuracy looks good! But some questions:

Will you compare the performance with non-tensorflow machine learning
algorithms such as boosted decision trees or support vector machines
(e.g. provided by scikit-learn or xgboost)?

Your model is a deep neural network, so there is a convolutional layer
in this model to make use of the 2D representation of the spectral
coherence function? DNN implies for me only a bunch of hidden layers. As
well, convolution layer(s) makes the classificator translation invariant
(do you need this?).

Greetings,
Stefan

On 06/29/2016 10:23 AM, Christopher Richardson wrote:
> Hi Everyone,
>
> I've just posted a blog article on what I've been working on
> recently - https://signalsintelligence.wordpress.com/2016/06/24/tensorflow/
>
> I'm now going to work on making a block for GNU Radio, which makes use
> of TensorFlow to classify signals.
>
> cheers
>
> Chris
>
>

> _______________________________________________
> Discuss-gnuradio mailing list
> address@hidden
> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
>