报告人：Moncef Gabbouj 院士

工作单位：坦佩雷大学（芬兰）

报告题目：用于多媒体数据分析和检索的机器学习和优化工具

报告时间：2023年4月12日（周三）14:30-15:30

报告链接：Teams Link

https://teams.microsoft.com/l/meetup-join/19%3aB4gmRcUATAMA2iJqi-xXvtfPFfTbxVJPxSW_pcAPBao1%40thread.tacv2/1638719716825?context=%7b%22Tid%22%3a%2222804ebb-30d5-47df-942f-f3a3722f0225%22%2c%22Oid%22%3a%2216a60c03-ad7a-4b85-a403-8ebd947e010c%22%7d 

内容摘要：

多媒体数据分析是现代决策环境的核心引擎。首先，本报告将讨论目前在多媒体数据分析中开发和使用的现代信号处理、机器学习、模式识别和优化工具，特别是多媒体数据检索，大型媒体库中的分类和搜索将是主要的目标应用。其次，本报告将涉及一种基于内容的多媒体搜索新范式。我们提出了一种用于信息检索的经典搜索引擎的替代方法，该方法可用于通用多媒体库，并且在一个集成网络（进化）二元分类器（CNBC）框架中引入了一个增量进化方案。该方案解决了特征/类的可扩展性问题，并在动态图像库中实现了高分类和基于内容的检索性能。CNBC成功的背后是实现CNBC主干的新颖设计，即二元分类器，这是一个特殊的神经网络，使用了最新提出的进化优化算法（多维粒子群优化）进行优化设计。粒子群优化（PSO）是基于种群的随机搜索和优化过程，由Kennedy和Eberhart于1995年提出。其目标是收敛到某些多维适应度函数的全局最优。本报告将介绍两种新技术，它们扩展了基本的PSO算法，即在一个最优维度未知的多维搜索空间中，种群粒子可以同时寻求位置和维度的最佳值，由此产生的MD-PSO在开发数据分类和特征合成的机器学习工具方面发挥了关键作用。目前，大多数基于内容的多媒体搜索引擎在很大程度上依赖于低级特征。然而，这种自动提取的特征通常缺乏准确描述图像内容所需的分辨能力，导致检索性能不佳。为了解决这个问题，我们提出了一种进化特征合成技术，该技术在优化选择的特征上寻求最优的线性和非线性操作，从而合成高辨识度的特征。其最优性是通过MD-PSO寻求的，合成的特征只需用于少数原始特征向量，并在不同类别之间表现出显著的区分能力，广泛的CBIR实验表明，该技术能够实现显著的性能改进。此外，本报告还将回顾并提出新的深度学习范式，该范式可能彻底改变媒体检索。

个人简介：

Moncef Gabbouj 院士于1985年在斯蒂尔沃特的俄克拉荷马州立大学获得电气工程学士学位，并分别于1986年和1989年在印第安纳州西拉法叶的普渡大学获得电气工程的硕士和博士学位。Moncef Gabbouj 院士现任芬兰坦佩雷大学计算机科学系信息技术教授、IEEE计算机科学协会哈里-古德奖委员会副主席，曾任芬兰学院教授以及不同大学的客座教授。Moncef Gabbouj 院士目前的研究领域包括大数据分析、基于多媒体内容的分析、索引和检索、人工智能、机器学习、模式识别、非线性信号和图像处理及分析、语音转换以及视频处理和编码。Moncef Gabbouj 院士目前是IEEE和亚太人工智能协会会员，欧洲科学院、芬兰科学和文学院以及芬兰工程科学院成员，IEEE CAS TC on DSP的前任主席以及IEEE信号处理协会傅里叶奖委员会成员，同时也是美国国家科学基金会IUCRC资助的视觉和决策信息学中心（CVDI）的芬兰站点主任，领导经济事务和就业部资助的自主系统研究联盟（RAAS）的人工智能研究任务组，曾任IEEE CASS特邀讲师，许多IEEE和国际期刊的副编辑和客座编辑，IEEE SPS和CAS旗舰会议、ICIP和ISCAS以及ICME 2021总主席。Moncef Gabbouj 院士曾获2017年芬兰文化基金会艺术与科学奖、2015年TUT基金会大奖、2012年诺基亚基金会客座教授奖、2005年诺基亚基金会表彰奖以及多个最佳论文奖，已出版两部专著和850多篇期刊和会议论文，指导了54篇博士论文和76篇硕士论文。

                                            【编辑：王健】

英文版：

Academic Report Notice of Moncef Gabbouj：Machine Learning and Optimization Tools for Multimedia Data Analytics and Retrieval

Speaker: Professor  Moncef  Gabbouj

Title: Machine Learning and Optimization Tools for Multimedia Data Analytics and Retrieval

Time: 14:30-15:30, April 12th, 2023 (Wednesday)

Website: Teams Link

https://teams.microsoft.com/l/meetup-join/19%3aB4gmRcUATAMA2iJqi-xXvtfPFfTbxVJPxSW_pcAPBao1%40thread.tacv2/1638719716825?context=%7b%22Tid%22%3a%2222804ebb-30d5-47df-942f-f3a3722f0225%22%2c%22Oid%22%3a%2216a60c03-ad7a-4b85-a403-8ebd947e010c%22%7d 

Abstract: 

    Multimedia Data analytics is the core engine in modern decision-making environments. In this talk, we discuss modern signal processing, machine learning, pattern recognition and optimization tools recently developed and used in multimedia data analytics with a special emphasis on multimedia data retrieval. Classification and search in large media repositories will be the main targeted applications. The talk deals with a new paradigm for multimedia search based on content. We present an alternative approach to classical search engines for information retrieval, which can be used for generic multimedia repositories. We introduce an incremental evolution scheme within a collective network of (evolutionary) binary classifier (CNBC) framework. The proposed framework addresses the problems of feature/class scalability and achieves high classification and content-based retrieval performances over dynamic image repositories. The secret behind the success of CNBC is a novel design to implement the backbone of CNBC, namely the binary classifier. This is a special neural network, which is optimally designed using the recently developed evolutionary optimization algorithm called multi-dimensional particle swarm optimization. Particle swarm optimization (PSO) is population based stochastic search and optimization process, which was introduced in 1995 by Kennedy and Eberhart. The goal is to converge to the global optimum of some multi-dimensional fitness function. Two novel techniques, which extend the basic PSO algorithm, are presented. In a multidimensional search space where the optimum dimension is unknown, swarm particles can seek both positional and dimensional optima. The resulting MD-PSO plays a key role in developing machine learning tools for data classification and feature synthesis. Most content-based multimedia search engines available today rely heavily on low-level features. However, such features extracted automatically usually lack discrimination power needed for accurate description of the image content and may lead to poor retrieval performance. To address this problem, we propose an evolutionary feature synthesis technique, which seeks for the optimal linear and non-linear operations over optimally selected features so as to synthesize highly discriminative features. The optimality therein is sought through MD-PSO. The synthesized features are applied over only a minority of the original feature vectors and exhibit a major discrimination power between different classes and extensive CBIR experiments show that a significant performance improvement can be achieved. The talk will also review and propose novel deep learning paradigms that has the potential to revolutionize media retrieval.

Personal Introduction:

     Moncef Gabbouj received his BS degree in electrical engineering in 1985 from Oklahoma State University, Stillwater, and his MS and PhD degrees in electrical engineering from Purdue University, West Lafayette, Indiana, in 1986 and 1989, respectively. Dr. Gabbouj is a Professor of Information Technology at the Department of Computing Sciences, Tampere University, Tampere, Finland. He was Academy of Finland Professor during 2011-2015. He held several visiting professorships at different universities. Dr. Gabbouj is currently the Finland Site Director of the NSF IUCRC funded Center for Visual and Decision Informatics. His research interests include Big Data analytics, multimedia content-based analysis, indexing and retrieval, artificial intelligence, machine learning, pattern recognition, nonlinear signal and image processing and analysis, voice conversion, and video processing and coding. Dr. Gabbouj is a Fellow of the IEEE and Asia-Pacific Artificial Intelligence Association. He is member of the Academia Europaea, the Finnish Academy of Science and Letters and the Finnish Academy of Engineering Sciences. He is the past Chairman of the IEEE CAS TC on DSP. He is a member of the IEEE Signal Processing Society Fourier Award Committee and Vice-Chair of the IEEE Computer Science Society Harry Goode Award Committee. He served as Distinguished Lecturer for the IEEE CASS. He served as associate editor and guest editor of many IEEE, and international journals as well as General Chair of IEEE SPS and CAS Flagship conferences, ICIP and ISCAS as well as ICME 2021. Dr. Gabbouj was the recipient of the 2017 Finnish Cultural Foundation for Art and Science Award, the 2015 TUT Foundation Grand Award, the 2012 Nokia Foundation Visiting Professor Award, the 2005 Nokia Foundation Recognition Award, and several Best Paper Awards. Dr. Gabbouj is the Finland Site Director of the NSF IUCRC funded Center for Visual and Decision Informatics (CVDI) and led the Artificial Intelligence Research Task Force of the Ministry of Economic Affairs and Employment funded Research Alliance on Autonomous Systems (RAAS). He published two books and over 850 journal and conference papers and supervised 54 doctoral and 76 Master theses. 

                                                                                                                                               [Editor:Jian Wang]