arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1809.04855stat.MLcs.LG
收藏

随机变分优化

Stochastic Variational Optimization

University College London(伦敦大学学院)

Thomas Bird, Julius Kunze, David Barber

摘要:

变分优化形成目标的可微上界。我们表明,自然进化策略和高斯扰动等方法,是变分优化的特殊情况下,其中的期望近似高斯采样。这些方法特别令人感兴趣,因为它们是可并行的。我们计算了相应的梯度估计的近似偏差和方差,并证明了使用对偶抽样或基线是至关重要的,以减轻他们的问题。我们将这些方法与另一种可并行化的方法,即方向导数进行对比。我们的结论是,对于可微目标,使用方向导数是最好的,使用变分优化进行并行随机梯度下降。

英文摘要:

Variational Optimization forms a differentiable upper bound on an objective. We show that approaches such as Natural Evolution Strategies and Gaussian Perturbation, are special cases of Variational Optimization in which the expectations are approximated by Gaussian sampling. These approaches are of particular interest because they are parallelizable. We calculate the approximate bias and variance of the corresponding gradient estimators and demonstrate that using antithetic sampling or a baseline is crucial to mitigate their problems. We contrast these methods with an alternative parallelizable method, namely Directional Derivatives. We conclude that, for differentiable objectives, using Directional Derivatives is preferable to using Variational Optimization to perform parallel Stochastic Gradient Descent.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑