Stochastic batch size for adaptive regularization in deep network optimization
作者：
Highlights：
• Adaptive regularization for deep network optimization via parameter-wise batch size.
• The stochastic batch size reflects local and global properties of each parameter.
• Beneficial for practical studies where the number of training examples is small.
摘要
•Adaptive regularization for deep network optimization via parameter-wise batch size.•The stochastic batch size reflects local and global properties of each parameter.•Beneficial for practical studies where the number of training examples is small.
论文关键词：Deep network optimization,Adaptive regularization,Stochastic gradient descent,Adaptive mini-batch size
论文评审过程：Received 4 August 2020, Revised 11 November 2021, Accepted 3 May 2022, Available online 5 May 2022, Version of Record 10 May 2022.
论文官网地址：https://doi.org/10.1016/j.patcog.2022.108776
