报告 人:张海斌教授(北京工业大学应用数理学院)
报告题目:NewDevelopment of Inertial Forward-Backward Algorithms
报告摘要:Withthe rapid development of large-scaledata, artificial intelligence, deep learning and other technologiesand applications, optimization algorithm design plays an increasinglyimportant role. In the field of signal and image processing, InertialForward-Backward Algorithms is given a lot of attentionbecause of its fast convergence rate in recent years.The classical methods is FISTA base on the Nesterov choice. For thedesign of fast optimization methods, the tuning of the sequence ofstep-sizes is a subtle issue. In this report, we show that theconvergence rate of the algorithm can be obtained simply by analyzingthe sequence of positive real numbers step-sizes, and, we unify mostof the existing results based on the accelerated gradient method ofNesterov.
报告时间:2019年4月8日(周一)上午10:00-11:30
报告地点:张衡楼106
主办单位:数学与统计学院、科研处、教务处
报告人简介:
张海斌,教授,博士生导师,运筹与科学计算系系主任,先后在美国明尼苏达大学、加拿大西蒙弗雷泽大学、澳大利亚迪肯大学做高级访问学者。主要从事最优化算法及其在数据处理等方面的应用,自动微分算法及其应用。先后主持国家自然科学基金面上项目3项:“变量部分稀疏正则化算法设计与研究”(11771003)、“信号和图像处理中的邻近点分裂算法研究”(61179033)和“自动微分实现与最优化方法研究”(10871014);发表学术论文40余篇、出版专著一部。