Generative Teaching Networks: Accelerating Neural Architecture Search by Learning to Generate Synthetic Training Data

If confirmed, then this is very promising and impressive research coming out of Uber. This new approach may speed up building and optimizing neural networks with less computational resources.

It is truly fascinating how much private companies (e.g. Amazon, Facebook, Salesforce, Google; Chinese companies etc.) and their many excellent researchers have contributed to the exponential and rapid progress and democratization of AI.

Generative Teaching Networks: Accelerating Neural Architecture Search by Learning to Generate Synthetic Training Data | Uber Engineering Blog