[nl]

Or you could just implement this (A submission from Google Brain to ICLR 2017):

In this paper, we use a recurrent network to generate the model descriptions of neural networks and train this RNN with reinforcement learning to maximize the expected accuracy of the generated architectures on a validation set. On the CIFAR-10 dataset, our method, starting from scratch, can design a novel network architecture that rivals the best human-invented architecture in terms of test set accuracy. Our CIFAR-10 model achieves a test error rate of 3.84, which is only 0.1 percent worse and 1.2x faster than the current state-of-the-art model. On the Penn Treebank dataset, our model can compose a novel recurrent cell that outperforms the widely-used LSTM cell, and other state-of-the-art baselines. Our cell achieves a test set perplexity of 62.4 on the Penn Treebank, which is 3.6 perplexity better than the previous state-of-the-art.[1]

To translate that, they built and train a RNN to design neural networks. These machine designed networks are almost equal to the best human designed network on an image-recognition benchmark, and outperform the best human-designed systems on a text understanding benchmark.

[1] http://openreview.net/forum?id=r1Ue8Hcxg