In this session we will enumerate and describe and discuss deep learning frameworks, platforms and libraries available now. A panel of experts will briefly describe leading frameworks. Then we will have an open discussion of their advantages and disadvantages and relationships to each other.
Newcomers may be unclear about options and which one to use. Even experts may benefit from a quick overview of the status of the different options especially if they have been working on one and have not had time recently to learn about new frameworks or to catch up on the others. We may all learn from any arguments or controversies or disagreements that arise.
Some examples of the kinds of questions we will discuss follow:
1) Since Keras runs on top of and wraps Tensorflow and sort of subsumes it, is there no longer any advantage to working directly with TensorFlow?
2) Or is there functionality that one can access in TensorFlow that is not accessible thru Keras? Also, TensorFlow has such strong support and is moving so rapidly…
3) is it possible that Tensorflow will roll over Keras by introducing the simplifications that make Keras desirable? Or
4) is the other main advantage of Keras, the ability to work on deep learning platforms other than TensorFlow, a strong enough advantage to ensure its survival?
5) What are the pros and cons of the BigDL library that scales deep learning using Spark and that takes advantage of math libraries that use Intel hardware architectures?
6) Suggest your questions in the discussion thread below, or at the session
BIO: Greg Makowski has been deploying data mining for 25 years, including ~25 “regular” (non-deep) neural networks, such as back-prop or Radial Basis Functions (RBF). He has deep learning experience with Convolutional Neural Networks applied to speech recognition, but not with the above frameworks.