These are chat archives for dmlc/mxnet

1st
Mar 2016
Jordan Micah Bennett
@JordanMicahBennett
Mar 01 2016 03:25
I now recognize of progression/'regression' amidst MXExecutorForward/Backward, ultimately abound libmxnet.dll
Jordan Micah Bennett
@JordanMicahBennett
Mar 01 2016 03:30
To conclude, par quantized regression, I must perturb mxnet c classes (low level mxnet mechanism--specifically elementwise_binaryop.cc with it's variants...)??
Jordan Micah Bennett
@JordanMicahBennett
Mar 01 2016 03:53
Since such shall solely manifest as gpu optmization, I shall then perhaps solely perturb cuDnn aligned propagation schemes/SHAPES, together with any cuDnn dependent inference/regression routine (via src/operator)?