VGG and Transfer Learning - Relationship to Greedy Layer-Wise Pretraining