Suppose, you have to design a deep convolutional netwrok based architecture for zooming or resizing images into larger one. Draw a block diagram for such a network. Also explain how you might collect/augment data for this problem and what type of loss function you might use. Also comment on the nature of the layers.

What are the challenges in training Generative Adversarial Networks? How are they mitigated in practice?

Explain the upsampling operation in convolutional neural networks. What are the applications for such operations?