Efficient Penalty-Based Bilevel Methods: Improved Analysis, Novel Updates, and Flatness Condition
高效罚函数法:改进分析、新更新方法及平坦性条件
机构 * University of Rochester(罗切斯特大学) ; Cornell University(康奈尔大学)
AI总结 本文提出PBGD-Free算法,通过解耦变量和引入平坦性条件,改进双层优化的收敛性和效率。
Comments arXiv admin note: text overlap with arXiv:2507.20400