报告题目 (Title)：Convex-Concave Minimax Optimization with the Second-Order Oracle Complexity of (*^(−4/7)) （凸-凹极小极大问题的复杂度为(*^(−4/7))的二阶算法）
报告人 (Speaker)：罗珞 教授（复旦大学）
报告时间 (Time)：2025年5月16日 (周五) 11:00
报告地点 (Place)：校本部GJ303
邀请人(Inviter)：徐姿 教授
报告摘要: The best-known second-order methods for solving convex-concave minimax problems require the second-order oracle complexity of O(ϵ^(-2/3) ), which has been speculated to be optimal. In this talk, we show an improved upper bound of O ̃(ϵ^(-4/7) ) by generalizing the optimal second-order method for convex optimization to solve the convex-concave minimax problem. The proposed algorithm also can be regarded as the second-order “Catalyst” framework for minimax optimization.