Deep Learning personal notes

1. For Gradient descent the initialization of weight could be anything including zero, since function is convex it will always converge to <global or local?> minima

2. How to check if the pytorch has mkl or not ?
>>> torch.has_mkl
True