scalnet

David did a great job with ScalaNLP. It happens to depend on Breeze. I would point out that Breeze does not support n-dimensional arrays (most tensors), although that is necessary to do in deep learning. We wrote ND4S and ScalNet to solve that: https://github.com/deeplearning4j/nd4s https://github.com/deeplearning4j/scalnet Moving computation out of Spark's MLlib and into lower level code like C++, as we do with JavaCPP and libnd4j, also improves speed. https://github.com/deeplearning4j/libnd4j https://github.com/bytedeco/javacpp

Deeplearning4j does this already. It has a huge community, a Scala API and does model import from Keras. It's important to note that Spark is not an efficient computation later -- it's best if used for fast ETL. If you get that wrong, your going to be training slow. https://deeplearning4j.org https://github.com/deeplearning4j/ScalNet https://deeplearning4j.org/model-import-keras https://gitter.im/deeplearning4j/deeplearning4j

Nah not our focus. We are going to interop with keras with our scala api: https://github.com/deeplearning4j/Scalnet That will allow us to benefit from what that community is building in terms of neural networks. We want to be the "deployment path" for those kinds of models while we focus on making c++ on the jvm as fast as possible. We basically focus on the hadoop and spark crowd. We're already the industry standard in java/scala: https://tech.knime.org/deeplearning4j https://www.rapidminerchina.com/en/products/shop/product/dee... https://www.lightbend.com/blog/meet-deeplearning4j-and-bidda... http://www.slideshare.net/agibsonccc/dl4j-in-the-wild