Here are the tasks we plan on working next. If you would like to contribute to one of them or have suggestions for more extensions you would like to see, please let us know!
- Switch from RDDs to DataFrames
- Make SparkNet a plug in replacement for MLlib's classification and regression algorithms
- Switch from images to arbitrary tensors
- Eliminate slowdown when Spark writes to disk
- Experiment with more communication schemes (for example ElasticSGD)
- Add examples for more models (for example RNNs)
- Use JavaCPP to wrap Caffe
- Wrap other deep learning libraries (TensorFlow, Torch)