Neural Architecture Search with Reinforcement Learning

TLDR

Neural networks are powerful and flexible models that excel at image, speech, and natural language tasks, yet designing them remains difficult. The authors aim to automatically generate neural‑network architectures by training a recurrent network with reinforcement learning to maximize validation accuracy. They employ a recurrent network that produces architecture descriptions and train it via reinforcement learning from scratch to produce novel network designs. On CIFAR‑10 the RL‑generated architecture attains a 3.65 % test error—0.09 % lower and 1.05× faster than the prior state‑of‑the‑art—while on Penn Treebank the learned recurrent cell achieves a perplexity of 62.4, 3.6 points better than the previous best, and sets a new state‑of‑the‑art character‑level perplexity of 1.214.

Abstract

Neural networks are powerful and flexible models that work well for many difficult learning tasks in image, speech and natural language understanding. Despite their success, neural networks are still hard to design. 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 3.65, which is 0.09 percent better and 1.05x faster than the previous state-of-the-art model that used a similar architectural scheme. 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 model. The cell can also be transferred to the character language modeling task on PTB and achieves a state-of-the-art perplexity of 1.214.