Message Boards Message Boards

Deep Learning with Wolfram Neural Networks?

GROUPS:

Is anyone interested in and actively using Wolfram to do Deep Learning research?

I am doing research in all areas of Deep Learning with my focus currently on VGG style of convolution neural networks (CNN). I will be branching out very soon to RLN, RNN, and GAN research very soon.

I would like to be able to compare experiences and lessons learned using Wolfram in these areas.

Thanks for any interest in discussions on Deep Learning.

POSTED BY: Bryan Minor
Answer
6 months ago

Just a note that while CNNs were already present in Wolfram Language 11.0:

https://www.wolfram.com/language/11/neural-networks

RNNs were added in Wolfram Language 11.1 and examples can be found in docs in:

http://reference.wolfram.com/language/ref/NetTrain.html

NetTrain >> Applications >> Sequence Learning and NLP

Please also see full info for available layers here:

http://reference.wolfram.com/language/guide/NeuralNetworks.html

POSTED BY: Vitaliy Kaurov
Answer
6 months ago

Thank Vitaliy for the information :)

I have doing numerous VGG CNN instances with 11.1 with success. Much to learn about the possibilities.

Do you know of any good examples with Wolfram Mathematica doing:

  1. RLN
  2. GAN Deep learning cases?
POSTED BY: Bryan Minor
Answer
6 months ago

I've never seen the acronym "RLN". Could you clarify what you mean by this?

Answer
6 months ago

Hi Sebastian,

Sorry if I miss spoke, I mean Reinforcement Learning Neural Network (RLNN).

The basic types we are working on are:

  1. CNN - Convolution NN
  2. Reinforcement Learning Neural Network (RLNN)
  3. RNN - Recurrent NN
  4. GAN - Generative Adversarial Networks

To date I have been working on variation of the VGG implementation of CNNs.

The other type I will be working on shortly, and at this point getting up to speed.

One further note, I am using Tensor Flow on also on a local GPU cube server, and on p2 blade instances at AWS.

POSTED BY: Bryan Minor
Answer
6 months ago

This is still not entirely clarified, as "RLNN" is still a non-standard acronym (as far as I know). Are you referring to policy networks in reinforcement learning? Perhaps send me a link to a paper or page which does what you want.

Answer
6 months ago

I am not an expert in RL. The paper of interest is: https://www.cs.toronto.edu/~vmnih/docs/dqn.pdf

POSTED BY: Bryan Minor
Answer
6 months ago

Sebastian -> I am thoroughly enjoying using Mathematica 11.1 to explore and learn on DL technology, thanks for your team's excellent work.

POSTED BY: Bryan Minor
Answer
6 months ago

Vitaliy,

Is the neural net code in Mathematica all "home grown" or do you use software such as tensorflow under the hood?

Regards

POSTED BY: Neil Singer
Answer
6 months ago

We use Apache MXNet (http://mxnet.io/) as a backend, and we also contribute improvements to this open source project.

Answer
6 months ago

Sebastian,

So the documentation of MXnet.io should be helpful for using DL in Mathematica 11.1?

POSTED BY: Bryan Minor
Answer
6 months ago

Group Abstract Group Abstract