Is there a visual tool for designing and applying neural nets/deep learning? I know there are lots of libraries for machine learning and deep learning like caffe, Theano, TensorFlow, keras, ...
But for me it seems like I have to know the architecture of the neural net, that I want to use.
Is there a (visual) tool that allows to experiment with different network designs and apply them on own data?
I'm thinking about something like the TensorFlow Playground, but with n-dimensional data and different layer types.
Thanks in advance!
 A: Yes,
There are many tools available for designing and applying neural network just by drag and drop.One of them is Deep Learning Studio Developed by Deep Cognition Inc, their robust deep learning platform with a visual interface in production provides a comprehensive solution to data ingestion, model development, training, deployment and management. Deep Learning Studio users have the ability to quickly develop and deploy deep learning solutions through robust integration with TensorFlow, MXNet and Keras.

Their auto ML feature will auto generate the neural network model.

A: For caffe there is a third-party tool called Expresso (http://val.serc.iisc.ernet.in/expresso/) that provides some GUI to help you getting started.
Moreover, NVIDIA DIGITS (https://developer.nvidia.com/digits) claims to be an interactive tool as well:

DIGITS simplifies common deep learning tasks such as managing data, designing and training neural networks on multi-GPU systems, monitoring performance in real time with advanced visualizations, and selecting the best performing model from the results browser for deployment. DIGITS is completely interactive so that data scientists can focus on designing and training networks rather than programming and debugging.

Hope this helps!
A: I've been working on a drag-and-drop neural network user interface (Ennui) that trains on the browser and allows users to export code-generated Python. We have various layers including dense, convolutional, maxpooling, batchnorm, etc. Building branched models like ResNets is also supported. We implemented a few common visualizations as well.  
Here is a picture of Ennui
Here is an example visualization 
You can visit the website at https://math.mit.edu/ennui
The open-source implementation is at https://github.com/martinjm97/ENNUI
Feel free to reach out with comments or questions. 
A: The process of finding the optimal network architecture for your problem is the heart of the deep learning process - that's where you use your prior knowledge to optimize performance. 
Honestly, I don't really see how a GUI as you suggested could serve this purpose, as:


*

*To be able to assess a given architecture, you need to train the net on your data (from scratch). For deep neural networks this is a process that could take a while. So if every click you make requires an hour's computation, it pretty much takes the entire advantage of a graphic UI off. 

*Most implementations (caffe, TensorFlow) have such simple syntax, that changing the architecture (changing up layers, tuning the hyper-parameters) really just comes down to changing the value of a single string or constant: nothing you really need a GUI for. 
If, on the other hand, what you are looking for is a more systematic approach to the parameter tuning business, you could read up on Automated Parameter Tuning.
A: Yes, there's a new visual editor for small neural networks called "Neural Network Designer" that is available at the Apple App Store for Mac.

