Stochastic Gradient Langevin Dynamics with Variance Reduction
Stochastic Gradient Langevin Dynamics with Variance Reduction
复制标题
DOI：
10.1109/ijcnn52387.2021.9533646
复制
发表时间：
2021-02
期刊：
影响因子：
--
通讯作者：
Zhishen Huang;Stephen Becker
中科院分区：
文献类型：
--
作者：
Zhishen Huang;Stephen Becker
随机梯度朗之万动力学（SGLD）因其全局寻优特性而受到了优化研究人员的关注。本文证明了利用方差约简加速的SGLD算法对非凸目标函数局部极小值的收敛性的改进。此外，我们证明了SGLD方案的遍历性，这给了它寻找非凸目标的全局最小值的潜力的见解。
Stochastic gradient Langevin dynamics (SGLD) has gained the attention of optimization researchers due to its global optimization properties. This paper proves an improved convergence property to local minimizers of nonconvex objective functions using SGLD accelerated by variance reductions. Moreover, we prove an ergodicity property of the SGLD scheme, which gives insights on its potential to find global minimizers of nonconvex objectives.