特邀报告

报告人:Maarten De Rijke

 Maarten de Rijke is full professor of Information Retrieval in the Informatics Institute at the University of Amsterdam. He holds MSc degrees in Philosophy and Mathematics (both cum laude), and a PhD in Theoretical Computer Science. He worked as a postdoc at CWI, before becoming a Warwick Research Fellow at the University of Warwick, UK. He joined the University of Amsterdam in 1998, and was appointed full professor in 2004. He is a member of the Royal Dutch Academy of Sciences (KNAW).
De Rijke leads the Information and Language Processing Systems group, one of the world’s leading academic research groups in information retrieval. His research focus is on intelligent information access, with projects on self-learning search engines, semantic search, and social media analytics.He is the director of Amsterdam Data Science. He’s a former director of the Intelligent Systems Lab (ISLA), of the Center for Creation, Content and Technology (CCCT), and of the University of Amsterdam’s Ad de Jonge Center for Intelligence and Security Studies.


报告题目:Neural Models for Modeling and Predicting Information Interaction Behavior
报告摘要:    Understanding user browsing behavior in web search is key to improving web search effectiveness. Many so-called click models have been proposed to explain or predict user clicks on search engine results. They are based on the probabilistic graphical model (PGM) framework, in which user behavior is represented as a sequence of observable and hidden events. The structure of the dependencies between the events has to be set manually. Different click models use different hand-crafted sets of dependencies.
We propose an alternative to the PGM-based approach, viz. a distributed representation-based approach, in which user behavior is represented as a sequence of vector states that capture the user’s information need and the information consumed by the user during search. These vector states can describe user behavior from more angles than the binary events used in PGM-based models (such as whether a user examined a document, or whether a user is attracted by a document), which makes them attractive for learning more complex patterns of user behavior than those hard-coded in existing click models.
We illustrate our approach using a set of neural click models. Our experimental results show that the neural click model that uses the same training data as traditional PGM-based click models, has better performance on the click prediction task (i.e., predicting user click on search engine results) and the relevance prediction task (i.e., ranking documents by their relevance to a query).
While click events are the most widely logged form of user interaction, there are also other behavioral signals that need to be understood, interpreted and modeled, and that can be used for ranking or prediction purposes. We show the versatility of our neural network-based approach to interaction modeling by capturing behavioral signals based on times between user actions. The ability to accurately predict(i) click dwell time (i.e., time spent by a user on the landing page of a search result), and (ii) times from submission of a query to the first/last click on the results and to the next query submission (if none of the results will be clicked) allows us to optimize search engines for constructing result pages and suggesting query reformulations that minimize time it takes users to satisfy their information needs.
In the final part of the talk I will discuss challenges that arise when we want to expand our neural network-based approach to interaction  modeling to deal with touch devices and with good abandonment.
The talk is based on joint work with Alexey Borisov, Aleksandr Chuklin, Julia Kiseleva, Ilya Markov, and Pavel Serdyukov.

报告人:梅俏竹

  梅俏竹,密歇根大学副教授。他广泛关注文本挖掘、信息检索、机器学习及其在网络搜索、社会计算和健康信息学中的应用。他是NSF CAREER奖以及ICML、KDD、WSDM、和其他顶级会议的多个最佳论文奖获得者。他是JMLR、TOIS、TWEB的编辑委员会成员,并且是在安娜堡举行的SIGIR2018的主席。


报告题目:Fighting Misinformation in Social Media through New IR Techniques
报告摘要: Rumors, or disputed factual claims, are widespread in social media which present serious threats to the social belief system. Identifying rumors is crucial in online social media where large amounts of information are easily spread across a large network by sources with unverified authority. It is extremely difficult to accurately classify whether every individual post is or is not making a disputed factual claim. In this talk, we will share our effort and vision of detecting, retrieving, analyzing, and predicting the impact of rumors in social media. Through tracking the signals that users enquiry the truth value of claims, trending rumors can be identified as early as possible in their diffusion. With an intelligent user-in-the-loop retrieval algorithm, posts that spread or correct a rumor in various ways can be identified with a high recall. Users can analyze the impact of the propagations and the corrections of a rumor through a new visualization tool for audience analysis. A carefully designed deep learning model can predict the influence of such cascades in their early stages.

报告人:马维英

  马维英博士,今日头条副总裁,人工智能实验室主任。他的主要研究领域包括机器学习, 自然语言处理, 多媒体分析和理解, 互联网搜索技术,知识图谱和数据挖掘。 加入今日头条前,马维英博士曾任微软亚洲研究院常务副院长,带领团队开发许多关键核心技术并用于微软必应搜索引擎Bing和在线广告系统,微软认知服务Cognitive Services,以及微软小冰聊天机器人和问答系统。他还将多项关键技术转移到微软的许多产品上, 包括Cortana, Exchange, SharePoint, Delve, Azure。他还领导团队在GitHub开源了多项尖端技术, 包括使得大规模机器学习任务具有高度可扩展性, 高效性和灵活性的分布式机器学习工具包Distributed Machine Learning Toolkit (DMTK), 和基于内存的分布式大规模图数据处理引擎Microsoft Graph Engine, 以及让计算机理解自然语言所需要掌握的概念和知识图谱Microsoft Concept Graph 。他本科毕业于台湾清华大学电气工程系,之后在美国加州大学圣芭芭拉分校深造,先后获得电气和计算机工程系(Electrical and Computer Engineering)硕士及博士学位。马维英博士是电气电子工程师学会院士(IEEE Fellow)、美国计算机协会杰出科学家(ACM Distinguished Scientist)、2008国际互联网大会(WWW)的程序委员会联合主席,以及2011年国际信息检索大会(SIGIR)的联合主席。他曾在界级会议和学报上发表论文270逾篇, 拥有160多项技术专利。


报告题目:智能连接人与信息: 信息流的未来与人工智能的机会
报告摘要: 如何更好地连接人与信息是人类社会的一个重要基础命题。在移动为先,万物互联,以及融合了社交的新内容时代,信息流成为一种新的连接方式。人工智能在这一领域里有着巨大的创新机会,帮助信息与内容的创作、 过滤、分发、搜索、消费以及互动几个环节提高效率乃至实现智能。演讲将会分享我对人工智能在此一基础领域的发展前景以及一些看法,包括人工智能的本质、近几年重要的技术发展,它对整个软件行业的影响以及企业应该如何建立人工智能的核心战略竞争力。另外我还将展示今日头条人工智能实验室的一些最新研究成果。

报告人:朱军

  朱军,清华大学计算机系长聘副教授、智能技术与系统国家重点实验室副主任、CMU兼职教授。01到09年获清华大学学士和博士学位,之后在CMU做博士后,11年回清华任教。主要从事机器学习基础理论和高效算法研究,在国际重要期刊与会议发表论文90余篇。担任IEEE TPAMI、Artificial Intelligence编委,ICML 2014地区联合主席, ICML (2014-2017)、NIPS (2013, 2015)、UAI (2014-2017)、IJCAI2015、AAAI(2016, 2017)等领域主席。获中国计算机学会青年科学家奖、国家优秀青年基金、中创软件人才奖等,入选国家“万人计划”青年拔尖人才、IEEE Intelligent Systems AI’s 10 to Watch、及清华大学221基础研究人才计划。


报告题目:大规模生成模型及应用
报告摘要: 生成模型是一类灵活的工具,可以从复杂数据中揭示隐含结构,生成新的样本。 在这个报告中,我讲介绍生成模型的大规模学习方面的进展,包括用于文本数据的主题模型, 用于图像的深度生成模型,以及“珠算”——一个支持概率编程和高效推理的Python库。

Tutorial

Tutorial 1:When game theory meets machine learning
主讲人:秦涛

秦涛博士,微软亚洲研究院主管研究员,中国科学技术大学兼职博导,于清华大学电子工程系获得学士和博士学位。他的主要研究领域包括机器学习和人工智能(重点是深度学习和强化学习的算法设计、理论分析及在实际问题中的应用),互联网搜索与计算广告,博弈论和多智能体系统。他在国际顶级会议和期刊上发表学术论文100余篇,曾任国际信息检索大会(SIGIR)、亚洲机器学习大会(ACML)、国际多智能体系统大会(AAMAS)领域主席,担任多个国际学术大会(NIPS、ICML、KDD、IJCAI、AAAI、WSDM、EC、SIGIR、AAMAS、WINE)程序委员会成员,曾任多个国际学术研讨会联合主席。



Tutorial 2:对话机器人的技术发展与挑战
主讲人:谭翊章

谭翊章(Wilson Tam)是微信人工智能(Wechat AI)的技术专家,专注对话机器人的技术研发。在2015-2017年期间,他担任微软苏州的资深自然语言处理科学家,参与微软小娜语音助手的技术研发。Wilson拥有6年在美国企业担任研究员的经验,包括SRI International Inc(Stanford Research Institute)和Nuance Communications Inc。在2009年,Wilson在美国Carnegie Mellon University获得计算机博士学位。他的研究兴趣包括语音识别、自然语言处理、机器翻译和机器学习等领域。