顶会顶刊论坛
顶会顶刊论坛报告1:
主讲嘉宾:韩军伟
时间:2021年9月26日9:40-10:00
地点:1楼经纬厅
韩军伟
报告题目:基于先进学习方法的协同显著性检测
报告摘要:图像和视频已经成为当今社会记录、表达和传递信息的主要媒介。人们希望通过计算机高效、准确地处理海量的图像视频数据。通过模拟人眼视觉注意机制来智能地检测显著性物体已经成为了计算机视觉领域中一个热门的课题,即显著性物体检测。近年来,为了进一步增加显著性检测的应用价值,以协同显著性检测为代表的一些新的研究分支开始出现并在领域内引起了广泛的研究兴趣。本报告将围绕协同显著性检测中的关键问题展开讨论,介绍基于多示例学习、度量学习、深度学习等先进学习方法的协同显著性检测框架(相关工作获国际期刊IEEE TCSVT 2021最佳论文奖),并讨论该研究领域具有潜力的未来研究方向。
个人简介:韩军伟,西北工业大学教授,获聘2018年度长江学者特聘教授,入选第四批国家“万人计划”科技创新领军人才,科睿唯安全球“高被引科学家”,爱思唯尔中国“高被引学者”。主要研究方向是人工智能、模式识别、类脑计算、遥感影像解译等。在领域顶级期刊/会议如:Proceedings of the IEEE,IEEE TPAMI,CVPR,ICCV,MICCAI等发表学术论文100余篇,论文被引用1.8万次,H-index 68。3篇论文入选年度中国百篇最具影响国际学术论文。获2021年度IEEE GRSS Highest Impact Paper Award(IEEE地球科学与遥感学会最有影响力论文奖)、国际期刊IEEE TCSVT 2021最佳论文奖、国际会议IEEE BIBM 2018最佳论文奖,国际会议ACM Multimedia 2010,MICCAI 2011和ICME 2016最佳学生论文奖提名。培养多名博士生/博士后获得中国图象图形学学会优秀博士论文奖、陕西省优秀博士论文奖、博士后创新人才支持计划等。获陕西省科学技术一等奖(排名第一)等6项省部级科技奖。担任IEEE Trans. on Neural Networks and Learning Systems、IEEE Trans. on Cybernetics、IEEE Trans. on Multimedia、《中国科学:信息科学》等多个国内外期刊编委,任国际会议如:CVPR, ICPR, ACCV等的领域主席。
顶会顶刊论坛报告2:
主讲嘉宾:Wu Liu
时间:2021年9月26日10:20-10:45
地点:1楼经纬厅
Wu Liu
报告题目:Progressive Search Paradigm for Open-world Instance Re-identification
报告摘要:The vast array of multimedia sensing technologies has produced a huge variety of big multi-modal data. The open-world instance re-identification (re-ID), whose primary tasks are finding a specific person/vehicle/object of interest with the multimedia sensing data, has tremendous potential applications. However, the open-world instance re-ID still faces the challenges of the guarantee of computing timeliness, the complex and instantly changed physical environments, the correlation discovery of massive multiple-modality data, and information security and privacy protection. To solve the challenges, we present a progressive search paradigm, which contains three important search strategies: 1) coarse-to-fine search in feature space; 2) near-to-distant search in spatial-temporal space; and 3) low-to-high permission search in the security space. The strategies all utilize simple features and computation to instantly reduce the search space, in which a complex matching process can be efficiently exploited to find the matched objects finely. Finally, we present our real-world produces implemented in the multimedia sensing network, which demonstrates the proposed progressive search paradigm can significantly improve the open-world Instance re-ID speed and accuracy.
个人简介:Dr. Wu Liu is a Senior Researcher in AI Research of JD.com, China. His current research interests include human behavior analysis and intelligent video surveillance. He received his Ph.D. degree from the Institute of Computing Technology, Chinese Academy of Science in 2015. He has published more than 80 papers in prestigious conferences and journals in computer vision and multimedia. He received IEEE Trans. on Multimedia 2019 Prize Paper Award, IEEE Multimedia 2018 Best Paper Award, IEEE ICME 2016 Best Student Paper Award, 2021 MSA-TC Best Paper Award-Honorable Mention, and Chinese Academy of Sciences Outstanding Ph.D. Thesis Award in 2016, etc. Dr. Liu is the founding committee member of ACM FCA, and the committee member of IEEE CASS-MSA. He has also served as the Technical Program Chair of ACM MM Asia 2021, Web Chair of ICME 2019, Publicity Chair of BIGMM 2018, Industrial Chairs of ChinaMM 2020&2021, and the Area Chairs of ACM MM 2019-2021, AAAI 2021, ICME 2019, ICIP 2017, etc. Wu Liu also organized the three workshops in ACM MM 2021&2020 and IEEE ICCV 2021, three tutorials in ACM MM 2020, IEEE ICME 2019 and ACM MM Asia 2019, four special issues in ACM TOMM 2021, IEEE TCSVT 2021, MVA 2018, and MTAP 2019.
顶会顶刊论坛报告3:
主讲嘉宾:仉尚航
时间:2021年9月26日10:45-11:10
地点:1楼经纬厅
仉尚航
报告题目:类Transformer稀疏注意力建模与长序列预测
报告摘要:长序列预测是人工智能研究的难点,对工业健康维护、疾病传播预测、网络安全分析等关键领域具有重要作用。本次报告将介绍获得AAAI 2021 最佳论文奖的工作“Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting”。该工作指出传统循环神经网络因误差逐层累积已不能满足长序列数据分析的需求,并首次明确了Transformer神经网络架构对长序列问题建模的重要意义。该架构的核心优势是构建了自注意力机制来捕捉跨长度的前后相关性,但其伴随的重大挑战是自注意力操作具有随输入长度的二次时间复杂度,无法适用于长序列输入和输出。Informer首次在长序列问题上运用Transformer神经网络架构,依靠所提出的可分析稀疏化、注意力蒸馏和生成式解码,可以在同样硬件限制下显著提高序列分析任务性能,为解决长序列问题提供了一种全新的解决方案。本次报告将介绍Informer如何突破传统自注意力的计算复杂度限制,提出一种全新的从概率角度进行自注意力矩阵稀疏化的模型ProbSparse Attention,并介绍Informer的注意力蒸馏机制和生成式解码。本报告包括传统序列模型的演进和发展,Transformer类模型在长序列预测的可能性,解析稀疏注意力模型,性能消融分析,展望未来长序列分析等。
个人简介:仉尚航,2018年博士毕业于美国卡内基梅隆大学 (CMU), 之后在美国AI科技公司Petuum担任研究科学家,后于2020年初加入加州大学伯克利分校 Berkeley AI Research Lab (BAIR)从事博士后研究。其针对智能终端泛化能力有限,依赖大规模标注数据,无法在开放动态环境中自动适应新数据、识别新事物的问题,进行强泛化机器学习理论算法研究,并对端云协作系统进行优化,提升泛化能力。在世界顶级期刊和会议上有30余篇论文发表,并申请5项美中专利。荣获世界人工智能顶级会议AAAI’2021 最佳论文奖,该工作开源近 3月以来代码访问 6 万余次、1500余次Star,并转化应用于国家电网。仉博士于2018年入选美国”EECS Rising Star”,曾获得Adobe学术合作基金,Qualcomm创新奖提名。作为编辑和作者出版英文书籍《深度强化学习》(Springer Nature), 至今电子版全球下载量超58000次,入选中国作者年度高影响力研究精选,并同时出版中文译本(电子工业出版社)。多次组织NeurIPS/ICML Workshop。
顶会顶刊论坛报告4:
主讲嘉宾:梁红茹
时间:2021年9月26日11:10-11:35
地点:1楼经纬厅
梁红茹
报告题目:PiRhDy: Learning Pitch-, Rhythm-, and Dynamics-aware Embeddings for Symbolic Music
报告摘要:Definitive embeddings remain a fundamental challenge of computational musicology for symbolic music in deep learning today. Analogous to natural language, music can be modeled as a sequence of tokens. This motivates most existing solutions to explore the utilization of word embedding models to build music embeddings. However, music differs from natural languages in two key aspects: 1) musical token is multi-faceted–it comprises of pitch, rhythm and dynamics information, and 2) musical context is two-dimensional, each musical token is dependent on both melodic and harmonic contexts. In this work, we provide a comprehensive solution by proposing a novel framework named PiRhDy that integrates pitch, rhythm, and dynamics information seamlessly. PiRhDy adopts a hierarchical strategy which can be decomposed into two steps: (1) token (i.e., note event) modeling, which separately represents pitch, rhythm, and dynamics and integrates them into a single token embedding; and (2) context modeling, which utilizes melodic and harmonic knowledge to train the token embedding. A thorough study was made on each component and sub-strategy of PiRhDy. We further validate our embeddings in three downstream tasks – melody completion, accompaniment suggestion, and genre classification. Results indicate a significant advancement of the neural approach towards symbolic music as well as PiRhDy’s potential as a pretrained tool for a broad range of symbolic music applications.
个人简介:梁红茹,四川大学计算机学院(软件学院)专职博士后,于2020年在南开大学计算机学院获博士学位,主要研究方向为计算音乐学、自然语言处理、表征学习等,相关研究成果发表在ACM MM、COLING等国际会议和期刊上,其中以第一作者发表在ACM MM 2020的长文被评选为“最佳论文”。
顶会顶刊论坛报告5:
主讲嘉宾:魏恺轩
时间:2021年9月26日11:35-12:00
地点:1楼经纬厅
魏恺轩
报告题目:噪声建模与图像重建
报告摘要:从不完整及低信噪比的观测数据中重建出清晰的目标图像是计算成像中一项根本的任务,其在遥感、生物光学、诊断医学等领域有着广泛的应用。图像重建结果很大程度上取决于所采用的噪声模型以及图像先验的准确性。高质量的重建结果依赖于精确的噪声及图像先验的建模。本次报告将着重介绍一种基于电子成像物理过程的噪声模型及一类通用的图像重建方法。从成像传感器的物理特性出发,精细地建模从光子到数字信号的电子成像物理过程中所涉及到的各类噪声源,并提出一种免调试即插即用的近端优化算法。突破传统成像系统的限制,在极暗光成像、压缩感知核磁共振成像、稀疏角度计算断层成像、相位恢复等任务中展示了优越的性能。
个人简介:魏恺轩,美国普林斯顿大学博士生,本硕分别于2018年及2021年毕业于北京理工大学。研究方向为计算机视觉、计算摄影学、计算成像学。以第一作者身份在IEEE TPAMI、TNNLS等国际期刊和CVPR、ICML等国际会议发表多篇论文,其ICML论文入选ICML 2020杰出论文奖,曾获2020年度百度奖学金,入选全球首份AI华人新星百强榜单,同时受邀担任多个主流国际会议及期刊的审稿人。
版权所有:中国计算机学会技术支持邮箱:conf_support@ccf.org.cn