while with high-performance you mean documented FFI and friends? this should be possible with R as well. It also seems that the actual open-source ML-community

while with high-performance you mean documented FFI and friends? this should be possible with R as well. It also seems that the actual open-source ML-community (vs. Google: we want you to use our software to ensure you can't ever own your stuff) supports R just fine: https://mxnet.apache.org/api/r https://mxnet.apache.org/api/r