The tests have been made for the following versions and the results have been generated only on these versions:
- Tensorflow 1.3.1 (Installed from source +CUDA + MKL + CUDNN)
- Keras 2.0.8 (Installed from pip + TF 1.3.1 as seen above)
- Caffe 1.0 (Installed from source +CUDA +MKL -CUDNN)
- Torch (3e9e141ced1afd0cad451e69f90e6e53503647ca) (Installed from source +CUDA +MKL)
- DL4J 0.9.1 (Using Maven + MKL in path)
Note: Experiment 3 and 4 are deprecated