-
Some New Directions in Online Structure Theory
Speaker: Rod Downey(professor in Victoria University of Wellington) Time: 10:00-11:00 (Time in Beijing) 15:00-16:00 (Time in Auckland) December 10, 2021 (Friday) Venue: B1-518B, Research Building 4 Abstract: I will report on some recent research giving a general framework for algorithmics on online structures. Currently there are many algorithms and no theoretical basis for this area. […]
-
Multilinear extension of k-submodular functions
Speaker: Baoxiang Wang(assistant professor in the Chinese University of Hong Kong) Time: 10:00-11:00 (Time in Beijing) 15:00-16:00 (Time in Auckland) December 3, 2021 (Friday) Venue: B1-518B, Research Building 4 Abstract: A -submodular function is a pairwise monotone function that given disjoint subsets outputs a value that is submodular in every orthant. In this paper, we […]
-
Introduction to Parallel Algorithms
Speaker: Yan Gu(professor in University of California, Riverside) Time: 10:20-11:20 (Time in Beijing) 15:20-16:20 (Time in Auckland) November 12, 2021 (Friday) Venue: B1-518B, Research Building 4 VooV Meeting ID: 359 812 755 Abstract: Parallel processors are ubiquitous nowadays and it is almost impossible to find a single-core processor, probably other than a toaster. However, very […]
-
On the reals weakly low for K
Speaker: Liang Yu(professor in Nanjing University) Time: 15:00-16:00 (Time in Beijing) 20:00-21:00 (Time in Auckland) November 01, 2021 (Monday) Venue: B1-518B, Research Building 4 Abstract: Given an infinite set , real is called weakly low for on if there are infinitely many so that does not improve the prefix-free complexity of up to a constant. […]
-
Randomness and Complexity
Speaker: Cristian Calude(professor in University of Auckland) Time: 14:00-15:00 (Time in Beijing) 18:00-19:00 (Time in Auckland) October 11, 2021 (Monday) Venue: Zoom Meeting ID: 711 8843 8437 Password: 202101 Abstract: Since ancient times randomness had been viewed as an obstacle and difficulty. This attitude has changed in the last century when randomness became central to […]
-
Rapid mixing of Glauber dynamics via spectral independence for all degrees
Speaker: Weiming Feng(research associate in University of Edinburgh) Time: 11:00-12:00 (Time in Beijing) 15:00-16:00 (Time in Auckland) September 17, 2021 (Friday) Venue: B1-518B, Research Building 4 Abstract: We prove an optimal lower bound on spectral gap of the Glauber dynamics for anti-ferromagnetic two-spin systems with vertices in the tree uniqueness regime. This spectral gap holds […]
-
Sparse Metric Repair and Distance Realease
Speaker: Chenglin Fan(Ph.D UT Dallas) Time: 09:00-10:00 (Time in Beijing) 13:00-14:00 (Time in Auckland) July 30, 2021 (Friday) Venue: VooV Meeting ID: 747 545 809 Abstract: Metric data plays an important role in various settings, for example, in metric-based indexing, clustering, classification, and approximation algorithms in general. Often such tasks require the data to be […]
-
面向多租户分布式机器学习的聚合传输协议
Speaker: Wenfei Wu(Tsinghua University) Time: 11:00-12:00 (Time in Beijing) 15:00-16:00 (Time in Auckland) July 16, 2021 (Friday) Venue: B1-501, Main Building Abstract: 随着机器学习数据集和模型的增大,机器学习的训练过程逐步被分布式部署到多服务器上,其中多worker向参数服务器PS交换梯度、更新模型的计算方式是一种典型的体系结构。但是,在这种体系结构下,PS容易成为通信瓶颈。我们设计了聚合传输协议ATP来解决这一瓶颈,同时支持在数据中心中的多租户多机柜部署。ATP利用最近的可编程交换机技术,将参数聚合的过程卸载到交换机上,从而减小了PS的网络流量和计算量。ATP协议包括交换机上的网内聚合计算服务、终端服务器的可靠传输、和高吞吐网卡的加速技术。我们将ATP对接PyTorch并在AlexNet、VGG等常用模型上进行测试,证明ATP能够有效的加速机器学习的效率。 Speaker Bio: 吴文斐,清华大学任助理教授。2015年博士毕业于美国威斯康星大学麦迪逊分校,后在惠普实验室任博士后研究院。2017年加入清华大学工作至今。SIGCOMM、NSDI、INFOCOM等网络顶级会议上发表论文30余篇,拥有美国专利3项。获SoCC13最佳学生论文、IPCCC最佳论文提名。 Download poster
-
启发式优化及其复杂工业应用
Speaker: Zhipeng Lv(Huazhong University of Science and Technology) Time: 11:00-12:00 (Time in Beijing) 15:00-16:00 (Time in Auckland) July 9, 2021 (Friday) Venue: B1-501, Main Building Abstract: 在各个领域的工业应用中往往存在大量的复杂优化问题,这些问题往往具有大规模、多约束、多目标、多层次、多维度、动态性强等特点,且往往被证明为NP完全或NP难的。学术界对优化问题的做法是“简单问题,复杂求解”(问题描述简单,但算法设计复杂),而工业界对优化问题往往是“复杂问题,简单求解”(问题描述复杂,但算法设计简单),而实际上对工业应用领域真正有价值的解决方案是对于复杂的工业问题使用高级的算法进行求解。因此需要建立一套从需求分析,到数学建模,到算法设计,再到算法工程的整体解决方案和技术途径。本报告将结合本团队在复杂工业优化领域的学术研究、算法竞赛、工业应用案例,重点讲解如何对复杂工业优化问题进行建模和求解,最后综述性地介绍求解复杂工业优化问题目前学术界主流的优化方法。 Speaker Bio: 吕志鹏,华中科技大学计算机学院教授,博士生导师,人工智能与优化研究所所长。主要研究方向为复杂系统建模、EDA算法、智能优化、调度与规划、启发式优化、NP难问题求解等。2012年入选教育部“新世纪优秀人才支持计划”。2008年获第二届国际大学排课表竞赛全球第二名,2010年获国际护士排班竞赛全球第三名,2016年获ROADEF/EURO液化气库存路由国际挑战赛全球第三名,2017年获SAT国际竞赛全球第一名,2018年获SAT国际竞赛季军全球第三名,2020年获GECCO会议最优摄像机布局竞赛三项全球第一名,2021年获ISPD会议物理建模“划分、布局和布线”算法竞赛全球第三名。在人工智能、计算机、运筹学、工业工程等领域的国际著名期刊和会议上发表学术论文70余篇(如AAAI, IJCAI, Artificial Intelligence, Transportation Science, INFORMS Journal on Computing等)。研究成果在航空、航天、通信、云计算、EDA、IC制造等领域得到应用,主持了二十余项大型企业应用优化项目,在复杂工业系统的智能优化方面为合作方提供了可供实用的解决方案。 Download poster
-
机器学习与量化投资
Speaker: Jian Li(Tsinghua University) Time: 09:30-10:30 (Time in Beijing) 13:30-14:30 (Time in Auckland) July 2, 2021 (Friday) Venue: B1-501, Main Building Abstract: 当代机器学习,尤其是深度学习的突破性进展已经给多个行业带来了巨大变革。将当代机器学习的工具与思想应用到量化投资领域是一个吸引人的课题,并得到了业界和学术界的广泛关注。量化投资是否可以归结为一个预测价格涨跌的监督学习问题?是否将传统的线性预测模型替换为最新得深度学习模型,我们就实现了机器学习与量化投资的结合?当代机器学习的工具到底能在哪些环节提高量化投资的能力与效率?在本次交流中我将以机器学习研究者的视角和大家对以上的问题进行交流,并分享我们团队近几年的思考和实践。 Speaker Bio: 李建,清华大学交叉信息研究院长聘副教授,博士生导师。他在中山大学取得的学士学位和复旦大学取得的硕士学位,马里兰大学博士毕业。他的研究兴趣主要包括算法设计与分析,机器学习,数据库,金融科技。他已经在主流国际会议和杂志上发表了70余篇论文等,并获得了VLDB 2009 和 ESA 2010 的最佳论文奖,ICDT201最佳新人奖,清华221基础研究青年人才支持计划,教育部新世纪人才支持计划,国家自然科学基金优秀青年基金。他主持并参与了多项科研项目,包括自然科学基金青年基金,面上项目,中以国际合作项目,青年973计划等。 Download poster