Low-Variance Gradient Estimation in Unrolled Computation Graphs with ES-Single
Vicol, Paul, Kolter, Zico, Swersky, Kevin
–arXiv.org Artificial Intelligence
We propose an evolution strategies-based algorithm for estimating gradients in unrolled computation graphs, called ES-Single. Similarly to the recently-proposed Persistent Evolution Strategies (PES), ES-Single is unbiased, and overcomes chaos arising from recursive function applications by smoothing the meta-loss landscape. ES-Single samples a single perturbation per particle, that is kept fixed over the course of an inner problem (e.g., perturbations are not re-sampled for each partial unroll). Compared to PES, ES-Single is simpler to implement and has lower variance: the variance of ES-Single is constant with respect to the number of truncated unrolls, removing a key barrier in applying ES to long inner problems using short truncations. We show that ES-Single is unbiased for quadratic inner problems, and demonstrate empirically that its variance can be substantially lower than that of PES. ES-Single consistently outperforms PES on a variety of tasks, including a synthetic benchmark task, hyperparameter optimization, training recurrent neural networks, and training learned optimizers.
arXiv.org Artificial Intelligence
Apr-21-2023
- Country:
- South America > Paraguay
- North America
- United States > California
- San Diego County > San Diego (0.04)
- Alameda County > Berkeley (0.04)
- Canada > Ontario
- Toronto (0.04)
- United States > California
- Europe > Germany
- Baden-Württemberg > Stuttgart Region > Stuttgart (0.04)
- Genre:
- Research Report (0.50)
- Technology: