Deeplearning4j Roadmap

These priorities have been set by what the Skymind has seen demand for among clients and open-source community members. Contributors are welcome to add features whose priority they deem to be higher.

High priority:

  • Parameter server
  • Sparse support for ND4J
  • Automatic differentiation
  • Python support/interface

Medium priority:

  • OpenCL for ND4J
  • CTC RNN (for speech etc.)

Nice to have:

  • Proper complex number support for ND4J (+optimizations)
  • Support for ensembles
  • Generative adversarial models

Low priority:

  • Hessian free optimization
  • Other RNN types: multi-dimensional; attention models, Neural Turing Machine, etc.
  • 3D CNNs

This is a work in progress. Last updated May 5, 2017.

Chat with us on Gitter