I dig hyper-parameter optimization and architecture discovery but I'm always reminded of [this slide](https://image.slidesharecdn.com/k2jeffdean-160609173832/95

I dig hyper-parameter optimization and architecture discovery but I'm always reminded of [this slide](https://image.slidesharecdn.com/k2jeffdean-160609173832/95/large-scale-deep-learning-with-tensorflow-46-638.jpg?cb=1465493958 https://image.slidesharecdn.com/k2jeffdean-160609173832/95/l...) and how these hp-tuning and architecture discovery tasks will always fall one tier above whatever your base optimization task is. The base tasks have to get faster for it to be practical for anyone without a server farm. I definitely think this is the future of ANNs, nets optimizing nets optimizing nets. Turtles all the way down.