https://hgpu.org/?p=17021
CHAOS: A Parallelization Scheme for Training Convolutional Neural Networks on Intel Xeon Phi