Understanding and correcting pathologies in the training of learned optimizers
This addresses a key bottleneck for researchers and practitioners in machine learning by making learned optimizers more practical, though it is incremental as it builds on existing methods.
The paper tackled the difficulty of training learned optimizers, which suffer from biased or exploding gradients, by proposing a dynamic weighting scheme for unbiased gradient estimators, enabling faster training of convolutional networks with improved test loss compared to tuned first-order methods.
Deep learning has shown that learned functions can dramatically outperform hand-designed functions on perceptual tasks. Analogously, this suggests that learned optimizers may similarly outperform current hand-designed optimizers, especially for specific problems. However, learned optimizers are notoriously difficult to train and have yet to demonstrate wall-clock speedups over hand-designed optimizers, and thus are rarely used in practice. Typically, learned optimizers are trained by truncated backpropagation through an unrolled optimization process resulting in gradients that are either strongly biased (for short truncations) or have exploding norm (for long truncations). In this work we propose a training scheme which overcomes both of these difficulties, by dynamically weighting two unbiased gradient estimators for a variational loss on optimizer performance, allowing us to train neural networks to perform optimization of a specific task faster than tuned first-order methods. We demonstrate these results on problems where our learned optimizer trains convolutional networks faster in wall-clock time compared to tuned first-order methods and with an improvement in test loss.