best365网页版登录学术报告

Generalized AFBA for saddle-point problems

白建超

（西北工业大学）

报告时间：2024年6月4日 星期二 下午16：00-17：00

报告地点：沙河校区E405

报告摘要：In this talk, we introduce a generalized asymmetric forward-backward-adjoint algorithm (G-AFBA) to solve a kind of saddle-point problems. Except enjoying proximal subproblems, G-AFBA enjoys more flexible and larger proximal stepsizes. We establish the global convergence and sublinear convergence rate of G-AFBA in the ergodic and pointwise senses. The linear convergence rate of G-AFBA is discussed briefly. We further discuss the application of G-AFBA to a multi-block separable convex optimization problem and a stochastic G-AFBA for solving a well-known machine learning problem. Experiments on testing the robust principal component analysis and the 3D CT reconstruction problems show that the proposed method outperforms several state-of-the-art methods.

报告人简介：白建超，博士 (后)，西北工业大学数学与统计学院副教授、陕西省优秀双创导师。他的研究兴趣包括数值代数与优化、机器学习和统计学习等领域的大规模优化方法、理论与应用，在Automatica、Computational Optimization and Applications、CSIAM Transactions on Applied Mathematics、IEEE Transactions on Medical Imaging、Journal of Scientific Computing、Numerical Linear Algebra with Applications等期刊上发表论文30余篇。曾主持1项国家级项目和5项省部级项目，现担任CSIAM大数据与人工智能专委会委员、陕西省科协人才奖励评审专家、陕西省/广东省/北京市/武汉市自然科学基金评审专家、PLOS ONE期刊编委等。

邀请人： 谢家新