旧版入口
学院动态
学术交流
2023年3月24日“优化计算及其应用”学术论坛
上传时间:2023-03-22 作者: 浏览次数:764

报告标题: 秩极小化的人体运动捕获数据恢复算法

报告人:赣南师范大学 胡文玉  

报告摘要:  Human motion capture (mocap) is the process of recording the information of key points of moving people through marker (e.g. Vicon) or markerless (e.g. MS Kinect) approaches. This technology is widely used to drive natural looking animations in a variety of applications such as movies, video games, virtual worlds, sports and medical treatment. However, the mocap acquisition process is not perfect, that is, some key points cannot be recorded due to occlusion, ambiguities, or other factors so that the acquired data usually has missing data. Thus, recovering the missing entries is a fundamental issue in mocap data analysis.

     By exploiting both the low-rank structure and temporal stability properties as well as the noise effect, the problem of mocap data recovery can be converted into a robust low-rank matrix completion (LRMC) problem whose model consists of three terms. To solve such a model, this report will introduce three new methods. The first one is derived from fixed-point proximity algorithm (FPPA) with convergence analysis. The second one is proposed by adopting truncated nuclear norm (TrNN) as an approximation to the rank of mocap data, and the last one is a discrete subspace structure (DSS) constrained approach which combines the LRMC and subspace clustering. Numerical experiments are performed to show their effectiveness and efficiency.

报告时间:2023324 (周五)上午8:30-12:00

报告地点:六教南528

报告人简介:胡文玉,博士、赣南师范大学数学与计算机科学学院教授、硕士生导师;主要从事机器学习与计算机视觉相关的数学理论、算法及应用研究,尤其在数据恢复、子空间聚类等方面获得了一些研究成果。先后访问了英国伯恩茅斯大学(国家计算机动画中心(NCCA))、中山大学、南开大学、成都电子科技大学等国内外知名学府。先后主持国家自然科学基金项目3项,江西省自然科学基金项目4项,在国际知名杂志,如:IEEE Trans. Image Process, Inform. Fusion, IEEE Signal Proc. Let., Neurocomputing, Appl. Math. Comput., Appl. Soft Comput.,  Appl. Math. Lett., J. Comput. Appl. Math., Theor. Comput. Sci. SCI期刊发表学术论文20余篇。到目前为止,指导硕士研究生11人,其中有2人应届考取博士研究生

 邀请人:喻高航


报告标题: A community-based centrality measure for identifying key nodes in multi-layer networks

报告人:中国计量大学 吕来水

报告摘要:  In this paper, we will define a community-based centrality for finding key vertexes in multi-layer networks, referred to as the CBCM. We first construct a multi-layer network model with connected edges between different network layers, which is represented by fourth-order tensor. Based on the fourth-order tensor, we develop a centrality, called PR BIS, to measure the importance of vertexes and network layers in multi-layer networks, simultaneously. CBCM determines the importance of a vertex in each network layer by combining the following three factors: the PageRank centrality score of the vertex, the importance of the community where the vertex is located, and the ability of the vertex within a community to affect vertexes in other communities within two steps. The importance of a community is determined by PageRank centrality value of the community, community modularity and community density. Based on the importance of all the network layers measured by PR_BIS centrality, we perform weighted fusion for the importance of a vertex in all network layers to obtain the importance of the vertex in multi-layer networks. To verify the effectiveness and superiority of CBCM, numerical experiments are performed on several multi-layer networks under the LT model.

报告时间:2023324 (周五)上午8:30-12:00

报告地点:六教南528

报告人简介:吕来水,2021年获得南京理工大学计算机科学与技术博士学位,目前就职于中国计量大学信息工程学院,研究方向为复杂网络理论与算法,在IEEE Transactions on Network Science and EngineeringApplied IntelligenceNeurocomputing国内外学术期刊发表SCI论文16篇,其中以第一作者发表SCI学术论文8篇,以通讯作者发表SCI论文2篇。

邀请人:喻高航


报告标题: 基于分布感知表示对齐的半监督域适应学习方法

报告人:北京航空航天大学 吴恒 

报告摘要:  域适应是一种在源域和目标域不满足独立同分布条件下的机器学习技术。相比同域的数据分布,域适应任务由于源域和目标域具有更难区分的外观信息差异,所以更有挑战性。针对域适应问题,本文在分布矫正的角度下,提出并实现了一种基于分布感知表示对齐的半监督域适应学习方法。首先,为了更有效的促进伪标签的生成,本文介绍了一种新颖的分布校准策略,通过将当前概率分布除以具有类均值的整体概率分布来控制错误伪标签的数量并提高伪标签的质量。在此基于上,进一步引入概率层和特征层表示对齐以减少域间的差异。其中,概率层表示对齐是混合源域的真实标签和无标签目标域的矫正标签作为预测器来监督混合的图像。特征层表示对齐是通过标签一致性选择模块查找源域和在无标签目标域中具有相同标签预测的实例,并强制相应的特征进行对齐,从而有效减少确认偏差和域偏移,提高源域到目标域的泛化能力。最后,在3个公开的半监督域适应数据集Office-31Office-homeDomainNet上进行了全面的实验。实验结果表明,本章所提出的方法超过了多种国际高水平的半监督域适应学习方法。消融分析和可视化也验证和解释了所提方法中不同模块的作用,并且有效促进域内特征和域间特征的对齐,提高模型在半监督域适应学习任务的性能。

报告时间:2023324 (周五)上午8:20-12:00

报告地点:六教南528

报告人简介:吴恒,男,北京航空航天大学计算机学院博士,2022年2月-2023年3月在深圳鹏城实验室进行访学。主要研究方向:计算机视觉、小样本学习、迁移学习、图像重建、医学图像处理等。在Knowledge-Based Systems (KBS)、Neurocomputing、Computational Visual Media (CVM)、IEEE International Conference on Multimedia & Expo (ICME)等国际期刊和会议上发表多篇论文。主持江西省校级研究生创新基金课题,并参与多项国家自然科学基金项目。

 邀请人:喻高航



学术报告

报告标题: 特征独立性和相关性关联学习的小样本图像分类方法

报告人:中国计量大学 郑子君

报告摘要:  小样本学习是计算机视觉中一项具有挑战性的任务,其目的是通过在已有知识的基础上来识别拥有稀缺样本的新类别。现有的小样本学习任务往往关注于特征的独立性学习,忽略了特征间的相关性学习。因此,在本文中提出了一种特征独立性和相关性关联关系的方法用于处理小样本图像分类问题。为了建立特征的独立性学习,本文通过数据增强监督每个实例的不同几何方向以及源域空间中实例的类别,从而增强特征的判别性。在此基础上,为了进行特征的相关性学习,本文利用了transformer单元构建全局和局部表示之间的长距离关系,并使用度量损失函数学习类内和类间变化,生成可靠的决策边界。通过联合独立性和相关性学习的关联学习操作,本方法可以显著提高模型对未见类别的泛化能力。在公开的小样本图像分类基准测试上进行的全面实验表明,我们提出的方法实现了最好的结果,并且验证了不同组成部分的有效性。

报告时间:2023324 (周五)上午8:30-12:00

报告地点:六教南528

报告人简介:郑子君,女,2022年毕业于华东理工大学信息科学与工程学院,获工学博士学位。现就职于中国计量大学3200威尼斯vip数据科学系。主要研究方向:人工智能、小样本视觉识别、时空大数据。在《Knowledge-Based Systems》、《Neurocomputing》、《Applied Intelligence》、《IEEE Transactions on Network Science and Engineering》等国际期刊上发表学术论文5篇,其中SCI收录5篇。获2019年江西省优秀硕士学位论文。作为主要成员参与科技部横向项目1项。

 邀请人:喻高航



报告标题: Singular Value Decomposition of Dual Matrices and its Application to Traveling Wave Identification in the Brain

报告人:复旦大学 丁维洋

报告摘要:  Matrix factorization in dual number algebra, a hypercomplex system, has been applied to kinematics, mechanisms, and other fields recently. We develop an approach to identify spatiotemporal patterns in the brain such as traveling waves using the singular value decomposition of dual matrices in this talk. Theoretically, we propose the compact dual singular value decomposition (CDSVD) of dual complex matrices with explicit expressions as well as a necessary and sufficient condition for its existence. Furthermore, based on the CDSVD, we report on the optimal solution to the best rank-k approximation under a newly defined Frobenius norm in dual complex number system. The CDSVD is also related to the dual Moore-Penrose generalized inverse.

Numerically, comparisons with other available algorithms are conducted, which indicate the less computational cost of our proposed CDSVD. Next, we employ experiments on simulated time-series data and a road monitoring video to demonstrate the beneficial effect of infinitesimal parts of dual matrices in spatiotemporal pattern identification. Finally, we apply this approach to the large-scale brain fMRI data and then identify three kinds of traveling waves, and further validate the consistency between our analytical results and the current knowledge of cerebral cortex function.

报告时间:2023324 (周五)下午14:00-16:30

报告地点:六教南528

报告人简介:丁维洋博士分别于2011年和2016年在复旦大学数学科学学院获得数学与应用数学专业的学士学位和计算数学专业的博士学位,他的博士导师是复旦大学的魏益民教授。2016年10月至2017年8月,他在中国香港理工大学应用数学系祁力群教授的团队作博士后研究。2017年9月至2020年11月,他在中国香港浸会大学数学系担任研究助理教授。其后于2020年11月加入复旦大学类脑智能科学与技术研究院,担任青年研究员。丁博士近期的主要研究兴趣包括结构矩阵、张量计算和优化及其在脑与类脑科学、数据分析、信号处理等领域的应用。

 邀请人:喻高航



报告标题: Frank-Wolfe type methods for a class of nonconvex inequality-constrained problems

报告人:浙江工业大学 曾燎原

报告摘要:  The Frank-Wolfe method and its variants, which implement efficient linear oracles for minimizing smooth functions over compact convex sets, form a prominent class of projection-free first-order methods. In this talk, we extend the Frank-Wolfe method and its away-step variant for minimizing a smooth function over a possibly nonconvex compact set, based on our new generalized linear oracles. We discuss convergence and present numerical performance of our nonconvex Frank-Wolfe type methods for solving matrix completion problems.

报告时间:2023324 (周五)下午14:00-16:30

报告地点:六教南528

报告人简介:曾燎原,于2018年在中国科学院数学与系统科学研究院获理学博士,其后于中国香港理工大学应用数学系从事博士后工作至2022年9月。现就职于浙江工业大学3200威尼斯vip数学系,任特聘副研究员。其主要研究求解大规模非凸优化问题的一阶算法,在SIAM Journal on Optimization, Computational Optimization and Applications 等期刊发表过多篇论文。

 邀请人:喻高航


Copyright © 2014 3200威尼斯vip(中国)有限公司 版权所有