Overparameterization creates a colossal hypothesis space. Although Golowich et al. (2018) have proved generalization bounds that have explicit independence on the network size, some implicit dependence may be observed.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Stochastic Gradient Descent as an Implicit Regularization

  • Fengxiang He,
  • Dacheng Tao

摘要

Overparameterization creates a colossal hypothesis space. Although Golowich et al. (2018) have proved generalization bounds that have explicit independence on the network size, some implicit dependence may be observed.