Google Brain Residency (#machine_learning)
▻http://tinyclouds.org/residency
From a software maintenance perspective there is little consensus on how to organize ML projects. It feels like websites before Rails came out: a bunch of random PHP scripts with an unholy mixture of business logic and markup sprinkled throughout. In TensorFlow projects it’s an unorganized glob of data pipelining, mathematics, and hyperparameter/configuration management. I believe a beautiful abstraction/organization has yet to be discovered.
[1702.00783] Pixel Recursive Super Resolution
▻https://arxiv.org/abs/1702.00783