Uber AI Labs has developed an algorithm called Generative Teaching Networks . that produces synthetic training data for neural networks which allows the networks to be trained faster than when using real data. Using this synthetic data, Uber sped up its neural architecture search deep-learning optimization process by 9x.
The CV system was able to achieve 98.9% accuracy with only 32 training steps. For a similar experiment on the CIFAR10 dataset, Uber showed that they could predict model performance with 128 training steps using synthetic data, compared to 1200 steps using real data, a speedup of 9x.