-add snOperatorCUDA project - everything related to CUDA ferried there
-refactoring
-speed in inference is not worse than Tensorflow

-add snSIMD project - AVX instr for CPU faster
-add operator 'Activation'
-add example resnet50 with weights from TF
-refactoring

-add AVX instr (CPU calc) for Convolution layer +20% performance
-add operator 'BatchNorm'
-refactoring
-fix bug

first release