第40期

题目：Iteratively Reweighted Methods for Solving Regularization Problems

主讲人：王浩，研究员

上海科技大学信息科学与技术学院

时间：2019年7月12日上午9:00

地点：海韵行政楼C510

报告摘要：This tutorialshall present two matrix-free methods for solving exact penalty subproblems on product sets that arise when solving large-scale optimization problems. The first approach is a novel iterative reweighting algorithm (IRWA), which iteratively minimizes quadratic models of relaxed subproblems while automatically updating a relaxation vector. The second approach is based on alternating direction augmented Lagrangian (ADMM) technology applied to our setting. The main computational costs of each algorithm are the repeated minimizations of convex quadratic functions which can be performed matrix-free. We prove that both algorithms are globally convergent un- der loose assumptions and that each requires at most O(1/ε2) iterations to reach ε-optimality of the objective function. We also extend our methods to solve Lp-norm regularized problems, which is nonconvex and nonsmooth.This tutorialpropose a general formulation of nonconvex regularization problems with convex set constraint, which can take into account most existing types of nonconvex regularization terms, bringing strong applicability to a wide range of applications. We design an algorithmic framework of iteratively reweighted algorithms for solving the proposed nonconvex regularization problems, which solves a sequence of weighted convex regularization problems with iteratively updated weights. We also provide global convergence under loose assumptions. This makes our method a tool for a family of various reweighted algorithms. The effectiveness and efficiency of our proposed formulation and the algorithms are demonstrated in numerical experiments for various regularization problems.

报告人简介：王浩博士于2015年4月在美国理海大学工业工程系获得博士学位，并于2010年和2007年在北京航空航天大学数学与应用数学系分别获得理学硕士和学士学位。在攻读博士期间，曾于2012年、2014年和2015年分别在埃克森美孚企业战略实验室、三菱电机研究实验室和群邑集团研发部担任实习研究员。当前研究领域主要为非线性优化、大规模优化、稀疏优化以及分布式优化。主要成果在优化顶级期刊SIAM Journal on Optimization上发表。

第41期

题目：Mobile Edge Artificial Intelligence: Opportunities and Challenges

主讲人：石远明，研究员

上海科技大学信息科学与技术学院

时间：2019年7月12日 上午10:30

地点：海韵行政楼C510

报告摘要：With the availability of massive data sets, high performance computing platforms, as well as sophisticated algorithms and software toolkits, artificial intelligence (AI) has achieved remarkable successes in many application domains, e.g., computer vision and natural language processing. However, AI tasks are computationally intensive and normally trained, developed, and deployed at data centers with custom-designed servers. Given the fast growth of intelligent devices, it is expected that a large number of high-stake applications (e.g., drones, autonomous cars, AR/VR) will be deployed at the edge of wireless networks in near future. As such, the intelligent wireless network will be designed to leverage advanced wireless communications and mobile computing technologies to support AI-enabled applications at various edge mobile devices with limited communication, computation, storage, hardware and energy resources. The aim of this tutorial is to present recent advances in mobile edge artificial intelligence systems, with a comprehensive coverage including edge inference process, edge training process and implementation issues. Through typical examples, the powerfulness of edge machine learning will be demonstrated, and their abilities in making low-latency, reliable and private intelligent decisions at network edge will be highlighted.

报告人简介：石远明博士于2011年7月获得清华大学电子工程学士学位（2007-2009学年于“数学物理基础科学班”培养）；2015年8月获得香港科技大学电子及计算机工程博士学位，师从Khaled B. Letaief教授。他于2015年9月加入上海科技大学信息科学与技术学院任助理教授/研究员，2019年1月任上海科技大学信息科学与技术学院常任副教授/研究员。他于2016年秋季学期任加州大学伯克利分校访问教授，访问统计机器学习领域Martin J. Wainwright教授。他的研究成果荣获2016 IEEE通信学会马可尼最佳论文奖(无线通信领域最重要学术奖项之一)，以及2016 IEEE信号处理学会最佳青年作者论文奖。他主要的研究方向为运筹优化、高维统计、机器学习、信号处理，及其在无线通信、量化金融中的应用。

邀请人：信息与通信工程系 付立群教授