第十届中国R会议（合肥）
分会场演讲摘要Ⅱ
生物与医疗专场
软件工具专场
机器学习与优化专场
生物与医疗专场
18日上午08:30-12:00
特种楼学术报告厅
“AI+慢性病管理”使
精准医疗成为可能
金博 大连理工大学 时间:8:30-9:15
主讲人简介
大连理工大学副教授。致力于数据挖掘、大数据分析、创新管理、商务智能等领域的科学研究。主持了国家自然科学基金青年项目、辽宁省高校科研项目、国家重点实验室开放课题等课题,参与科技部国家重点研发计划“精准医疗研究”项目、国家自然科学基金重大研究计划培育项目和面上项目、863计划项目等国家级课题。在相关领域重要国际期刊及会议上发表论文60余篇,近年来多篇论文在数据挖掘领域顶级会议(KDD、AAAI、ICDM、SDM、PAKDD等)收录,担任数据挖掘领域三大顶级会议KDD、ICDM、SDM的程序委员,是ACM、IEEE和CCF高级会员。
报告摘要
调查显示,慢性病及其并发症的急性发作已成为威胁我国老年患者健康的最主要因素。以帕金森症、阿兹海默症等神经系统慢性退行性疾病为研究对象,针对临床医学研究中的慢性病并发症评估、药品不良反应预测、联合用药推荐等难题,采用机器学习和医疗大数据分析的方法,在前期积累的海量医疗数据基础上,构建人工智能+慢性病管理的模式,以数据为驱动,使精准医疗成为可能,为提高我国医疗信息服务水平、合理利用医疗资源、探索新的慢性病并发症个性化治疗模式提供理论与实践支撑。
Smart Monitoring for Complex Diseases by Collaborative Learning and Selective Sensing
黄帅 华盛顿大学 时间:9:15-10:00
主讲人简介
Dr. Shuai Huang is an Assistant Professor at the Department of Industrial and Systems Engineering at the University of Washington. He received a B.S. degree on Statistics from the University of Science and Technology of China in 2007 and a Ph.D. degree on Industrial Engineering from the Arizona State University in 2012. He is also an adjunct faculty member at the Department of Biomedical Informatics and Medical Education (BIME) and the Integrated Brain Imaging Center (IBIC) at the University of Washington. Dr. Huang develops methodologies for modeling, monitoring, diagnosis, and prognosis of complex networked systems such as the brain connectivity networks, manufacturing systems, and disease progression process of complex diseases that have multiple stages and pathways. He also develops statistical and data mining models to integrate massive and heterogeneous datasets such as neuroimaging, genomics, proteomics, laboratory tests, demographics, and clinical variables, for facilitating scienti c discoveries in biomedical research and better decision-makings in clinical practices. His research is funded by the National Science Foundation, National Institute of Health, Juvenile Diabetes Research Foundation, Helmsley Foundation, and several biomedical research institutes. Dr. Huang currently serves as Associate Editor for the IIE Transactions in Healthcare Systems Engineering and Quality Technology and Quantitative Management.
报告摘要
The emerging data-rich environments in healthcare hold great promises to accelerate the paradigm transition of U.S. healthcare from reactive care to preventive care. One question is how we could translate the big disease data into better care management of preclinical or diseased patients. While these diseases manifest complex progression process, involving both temporal dynamics and spatial evolution, how could we model, monitor, and modify these processes are challenging problems. The challenges mainly lie on three aspects: disease modeling, monitoring, and prognosis. For example, diseases such as Alzheimer’s disease and Type 1 Diabetes share the commonality that they involve slow and predictable progression processes. Knowing how a disease progresses is helpful, particularly if we’d like to prevent the disease as early as we could for maximum therapeutic e cacy and improved quality of life. The modeling of the progression process is statistically chal- lenging given the high-dimensionality of the data (e.g., tens of thousands variables), the mixed types variables, and the data’s longitudinal nature. Another commonality of these diseases is that, since they are chronic condi- tions, being able to recognize subtle symptoms that indicate signi cant clinical events or suggest worse outcomes is crucial for preventative care. Further, patients need to be dynamically prioritized by their projected risk for resource allocation optimization. This needs robust models that build on the statistical knowledge provided by disease modeling and monitoring, to guide the selection of high-risk patients for targeted care. Thus, my works collectively work towards the goal of smart monitoring. Such a smart monitoring method will provide data-driven decision-making capabilities for better disease management, leading to e cient targeted screening and a ordable care, better treatment planning, and improved quality of life for both patients and caregivers.
Detecting concordance and discordance changes among a series of large-scale data sets
赖颖蕾 乔治华盛顿大学 时间:10:30-11:15
主讲人简介
Dr. Yinglei Lai is Professor of Statistics at The George Washington University. His research interest is to develop statistical and computational methods in bioinformatics, computational biology and biostatistics. He received his B.S. in Information & Computation Sciences and Business Administration from the University of Science and Technology of China in 1999. Dr. Lai received his Ph.D. in Applied Mathematics (Computational Biology) from the University of Southern California in 2003. After his postdoctoral training at Yale University School of Medicine, he joined as a faculty member in the Department of Statistics at the George Washington University in 2004.
报告摘要
With the current microarray and RNA sequencing technologies, two-sample genome-wide expression data have been increasingly collected in biological and medical studies. Di erential expression analysis and gene set enrichment analysis have been frequently conducted. The related statistical software in R has been widely used. Integrative analysis can be conducted when multiple data sets are available. In practice, concordant and discordant molecular behaviors among a series of data sets can be of biological and clinical interest. There is still a lack of statistical methods and software for these types of integrative analysis.
We have proposed a mixture model based approach to the integrative analysis of multiple large-scale two- sample expression data sets. Since the mixture model is based on the transformed di erential expression test P-values (z-scores), it is generally applicable to the expression data generated by either microarray or RNA sequencing platforms. The mixture model is simple with three normal distribution components for each data set to represent down-regulation, up-regulation and no di erential expression. However, when the number of data sets increases, the model parameter space increases exponentially due to the component combination from di erent data sets. To achieve a concordant and discordant integrative analysis for a series of data sets, We have introduced two model reduction strategies. The related statistical computing has been implemented in R.
We demonstrate our methods on the recent TCGA RNA sequencing data. To illustrate a concordant integrative analysis, we apply our method to a series of data sets collected for studying two closely related types of cancer. To illustrate a discordant integrative analysis, we apply our method to a series of data sets collected for studying di erent types of cancer. Interesting disease-related pathways can be detected by our integrative analysis approach.
眼底图像自动识别与诊断
蒋宇康 中山大学 时间:11:15-12:00
主讲人简介
中山大学数学学院统计学本科生,华南统计研究中心和R Square成员,对数据分析与R语言有着浓厚的兴趣。在华南统计研究中心的学习工作中,接触并分析过用户收视数据、基因数据、生存数据、图像数据等。
报告摘要
医学图像处理是当今非常热门的一个话题。其中包括CT、MRI等三维图像重构,多种疾病的诊断,病变部位标记等等。现阶段,医学眼底图片需要专业的医生来读图诊断,并且标记各种眼部疾病,需要大量的人力,花费大量的时间。基于这样的现状,我们通过自动化诊断系统诊断眼底疾病,节省医生时间,提高工作效率。本次分享将介绍,从眼科医生的实际需求出发,如何自动对视盘、视杯与黄斑进行识别;对血管静动脉进行标定;对出血、渗出、微血管瘤进行检测。
软件工具专场
17日下午14:00-17:30
西活三楼多功能厅
R 与深度学习的应用
李舰 微家实业 时间:14:00-14:45
主讲人简介
现任微家实业CDO,“统计之都”核心成员之一,曾任Mango Solutions中国区数据总监。台湾辅仁大学博士生在读,北京大学软件工程硕士,中国人民大学统计学学士。专注于数据科学在行业里的应用,著有《数据科学中的R语言》一书,曾在华东师范大学、浙江大学等高校任兼职导师,讲授数据科学相关的专业课程。
报告摘要
近年来,随着人工智能在图像识别、围棋竞技等领域的突破,带来了一波投资热潮,也使得深度学习变得火热。此外,TensorFlow、MXNet等优秀工具的成熟也使得深度学习的门槛一再降低,在很多领域都实现了成功的应用。本次报告试图用一种最简单的方式介绍深度学习技术的来龙去脉和实现原理,并分享图像识别的应用案例。此外还会介绍使用主流工具进行实际操作的方式,并对未来的可能发展方向进行展望。
Detection and Tracking
陈天龙 中国科学技术大学 时间:14:45-15:30
主讲人简介
Mathematics and Applied Mathematics and Computer Science double degree in USTC.
Research Experiences:
International Genetically Engineered Machine Competition Golden Prize;
Software Development and Deeping learning in Bioinformatics, Biomedical Cybernetics Laboratory, Harvard;
Elastic Optical Networks, Large Scale Computingand Networking Lab, USTC ;
Classi?cation of skin diseases based on image data, Graphics &Geometric Computing Lab, USTC .
报告摘要
The main thing I am going to show is a Detection and Tracking framework, which realizes capturing object from a video and tracking it. Firstly, I detect object at the start frame by using methods of RCNN, SPP, Fast-RCNN, Faster-RCNN, YOLO, SSD and YOLO-V2. Then, the bounding- box of the previous result it conveyed to the Tracking algorithm. I mainly focus my research on KCF and MDnet, by which the tracking performance of the object is tested. One thing need mentioning is that after a certain period of time, the location of the object is corrected using the former detection algorithm.
Great Again or Stronger Together? Sentiment Analysis About Book Reviews on Amazon
黎思言 中央财经大学 时间:15:50-16:35
主讲人简介
中央财经大学应用统计专业硕士,中国人民大学五校联合培养大数据分析硕士。参与《基于出租车数据分析北京市城市道路支路利用率》,聚类分析医学类微信公众号推文,“A Combination of CNN and VADER For Sentiment Classi cation”等项目。
报告摘要
我们收集了希拉里和特朗普在2016年美国大选期间发表的两本畅销书在亚马逊上的所有书评。然后使用情感分析和文本分类方法对两本书的书评进行了分析。我们得到的信息有:在亚马逊的书评中,对特朗普畅销书的评价呈现出积极的情感特征,书评内容和美国未来发展紧密相关。相反,对希拉里畅销书的评价呈现出消极的情感特征。
Consistent Multiple Change-point Detection and R implementation
李亚光 中国科学技术大学 时间:16:35-17:00
主讲人简介
中国科学技术大学统计金融系博士研究生,2012年毕业于华南理工大学,获理学学士学位。感兴趣的研究领域有:高维数据分析,change-point,变量选择等。目前的研究课题集中在高维数据的结构识别,尤其是高维数据的变点检测问题。
报告摘要
在变点分析中,对特定时间序列的多变点检测问题一直是统计学中的热点问题。R语言是从事统计研究的必备工具。本次报告将介绍常见的多变点检测方法和常用的R package,特别的本次报告还将介绍对一类非平稳时间序列模型以及线性回归模型的多变点检测方法及其R语言实现。最后通过模拟数据和实例说明该检测方法的有效性。
高频金融数据的非参数分析方法
徐刚 中国科学技术大学 时间:17:00-17:30
主讲人简介
中国科学技术大学统计与金融系硕士研究生在读，2014年毕业于中国科学技术大学少年班学院，获统计学学士学位。研究兴趣为非参数统计，小波分析，金融数据分析等方向。
报告摘要
金融数据时常会有不连续的跳跃点，很多时候这些跳跃点是不可忽略的。因此分析带跳跃的高频金融时间数据相比于传统的连续过程有显著不同。本报告主要介绍利用核方法和小波方法在跳跃点未知时识别这些跳跃点，并对含有跳跃点的高频金融降噪。
机器学习与优化专场
18日下午14:00-17:30
西活多功能厅
Targeted Sampling and Pricing Strategy with Imperfect Targetability and Customer Learning
邓世名 华中科技大学 时间:14:00-14:45
主讲人简介
邓世名博士,华中科技大学管理学院,教授,博士生导师.楚天特聘学者,入选教育部新世纪优秀人才计划,武汉市东湖高新科技开发区3551人才支持计划,湖北省“百人计划”。北京大学物理学士,计算机辅修学士。2003年获得美国加加利福尼亚大学伯克立分校工业工程与运筹学系博士学位。多年从事管理科学和运营管理中的研究和实践。曾在美国国际企业规划管理系统软件界领先的Oracle总部工作。在定价和制造联合决策,投资成本避税理论,供应链风险管理和合同设计理论,半导体可重组生产装备的产能规划研究等多项领域都有创新性的研究成果。
报告摘要
We investigate how sellers should design targeted sampling and pricing strategy when consumers have product t uncertainty and/or the targeting method cannot perfectly identify customers with speci c traits. We consider three cases: (1) perfect targetability and perfect learning; (2) imperfect targetability; and (3) imperfect learning. We reveals that the decisions of targeted pricing and targeted sampling are a ected by each other. With the presence of imperfect targetability, a over-targeting problem of choosing too many attributes in grouping consumers may occur. The consumer grouping decisions depends on both the degree of targetability and consumers’ability to learning product with provided samples.
DataBrain,
基于 R 语言开发的机器学习引擎
杨滔 桃树科技 时间:14:45-15:30
主讲人简介
杭州桃树科技有限公司创始人及首席执行官,拥有超过十年机器学习技术研究与应用经验。奥克兰大学机器学习博士,悉尼科技大学博士后。曾任阿里巴巴集团数据科学家,建立淘宝网数据科学团队,首创聚划算爆款模型。曾任F团首席科学家,建立F团数据化运营体系。
报告摘要
桃树科技产品DataBrain是一款机器学习引擎。DataBrain为数据分析师、数据工程师、信息工程师以及数据科学家提供端到端的数据建模工具。基于DataBrain所产生的信用风控、精准营销、个性化推荐和定投策略等模型已经嵌入多家银行、券商、互联网公司和IT企业的业务流程,大幅度提高企业运营效率。DataBrain降低机器学习和人工智能门槛,提供端到端的建模工具,让人人都可以成为数据科学家。
First Order Methods for Fast Linear Programming in SHUFE
邓琪 上海财经 时间:16:00-16:45
主讲人简介
博士,上海财经大学信息管理与工程学院讲师,博士毕业于美国佛罗里达大学,本科毕业于上海交通大学,运筹优化与机器学习领域青年研究学者。主要研究方向为运筹优化,以及优化算法在深度学习中的应用,有若干篇学术论文在机器学习及可视化领域顶级会议、期刊发表。目前参与信管学院、交叉学院与管科中心机器学习与并行优化软件开发。
报告摘要
线性规划是数学优化的核心问题之一,在管理科学和数据科学中有极其重要的作用,并被广泛而深入地研究。在求解线性规划的问题上,虽然传统的方法(单纯形法或内点法)能获得很高的精度,但它们并不能有效推广到大数据问题。本发言将介绍在大数据时代线性规划算法设计的挑战,以及基于一阶算法的线性规划求解方案。
基于低秩近似的一般性
增量矩阵分解框架
黄训蓬 中国科学技术大学 时间:16:45-17:30
主讲人简介
中国科学技术大学在读研究生。主要研究方向为增量学习与随机优化，着重于使用低秩近似方法解决高维优化变量运算时的时空复杂性问题。以第一作者主持的工作：“Incremental Matrix Factorization: A Linear Feature Transformation Perspective”已被IJCAI-17 接收，另有2篇论文于ICDM-17 、Information Science在投。
报告摘要
矩阵分解作为协同过滤中最常用的技术之一，有着高效、易于实现等一系列优点。但是由于矩阵分解本身是批量式的模型，如何在在线场景中随着已知元素的增加，增量地更新潜在特征矩阵是一个亟待解决的问题。目前大部分已有增量矩阵学习方法都有着较大的局限性，或是针对某些特定矩阵分解模型进行设计，或是有着较强的使用条件限制。在该报告中我们将介绍了一种一般性的增量矩阵分解框架，从低秩近似的角度对该方法进行了解释，给出了某些特殊情况下增量学习部分训练误差的上界。此外在多个数据集上进行的实验，验证了该方法从数据存储角度实现增量更新的同时保证了分解结果的高效与精确。
线上直播
本次会议将进行全程同步直播，诚邀大家在线观看！
或者点击网址：https://www.itdks.com/dakashuo/playback/935
联系方式
组委会主席：林枫
会议主页： https://china-r.org/hefei2017/index.html
会议微信：统计之都
新浪微博：@统计之都
微信公众号：统计之都
邮箱：chinar-2017-hf@cos.name
主办方
协办方
战略合作伙伴
直播支持
统计之都：专业、人本、正直的中国统计学社区。
往期推送：进入统计之都会话窗口，点击右上角小人图标，查看历史消息即可。
