eMatrix

微软亚洲研究院通讯

eMatrix

MSR Asia PR Newsletter

2017年07月

2017 July

中文  |  EN

焦点纵横

  • 微软AI发布会,集齐六大亮点召唤黑科技!

    7月12日,微软在英国伦敦举办了一场关于人工智能的发布会,一系列全新的项目和产品精彩登场,其中亮点包括:1. 宣布成立微软研究院人工智能中心;2. 发布全新人工智能设计指南;3. 宣布进行“地球人工智能计划”,用人工智能解决环境问题;4. 发布Seeing AI,帮助盲人和低视力群体开辟视觉世界;5. 宣布推出Presentation Translator演讲实时翻译字幕功能;6. 展示产品应用场景,人工智能可以多领域渗透生活。

  • 看微软亚洲研究院如何识人、用人、育人

    为了解答读者关于“如何才能加入微软亚洲研究院?”的问询,近期微软亚洲研究院特别推出了“科学匠人”系列文章,讲述研究院里典型与非典型研究员的成长故事,揭秘加入研究院的最佳途径。首篇文章阐明了微软亚洲研究院的人才观——始终坚持刚柔并济的原则:有高质量的标准,同时也有很大的自由度;不用死板的KPI作为考核绩效的标准,而是根据所做项目的影响力来进行综合衡量。在这里,每一个小创新,都会得到认可。

  • 微软推出演讲实时翻译字幕功能 Presentation Translator

    近日,Microsoft Translator通过微软车库项目发布了一项可显示演讲实时翻译字幕的最新功能——Presentation Translator。该功能以PowerPoint插件的形式呈现,可在幻灯片播放的过程中实时地显示翻译字幕,目前可以支持超过60种语言的字幕。不仅如此,观众们还可以通过Microsoft Translator APP加入会议,在智能设备或电脑上看到各自选定的语言字幕。

Spotlight

  • Six highlights announced at Microsoft London AI event

    On July 12nd, Microsoft hosted an AI thought leadership event in London. Harry Shum, Microsoft EVP, led the presentation, and key Microsoft R&D leaders shared Microsoft’s progress and plans in the AI space. Here are the six highlights: 1. Announcement of Microsoft Research AI; 2. Release of the Ethical Design Guide for AI; 3. Announcement of “AI for Earth”; 4. Release of Seeing AI; 5. Release of Presentation Translator; 6. Demonstration of various customer scenarios using Microsoft’s AI products.

  • How MSR Asia identifies, uses and fosters talents

    To answer readers’ questions about how to join MSR Asia, MSR Asia AI Headlines launched a special column called “Science Master” to share researchers’ stories of growth. The first article illustrated MSR Asia’s talent philosophy: MSR Asia has high-quality standards but also a lot of freedom; the only measure of performance is the impact of the projects, rather than KPI. Every little innovation is recognized at MSR Asia. Adhering to the principle of diversity and inclusiveness, MSR Asia always encourages cross-border innovation in interdisciplinary areas.

  • Presentation Translator, a Microsoft Garage project, provides translated subtitles during live presentations

    Recently, Microsoft Translator launched Presentation Translator, a Microsoft Garage project. This PowerPoint add-in provides translated subtitles during live presentations, displaying the subtitles directly on a presentation in any of more than 60 supported text languages. In addition to translated subtitles, audience members in the room can follow along with the presentation in their language of choice using their own phone, tablet, or computer.

媒体视角

近日,ZDNet Korea对微软全球资深副总裁、微软亚太研发集团主席兼微软亚洲研究院院长洪小文进行了书面采访。文章充分阐述了,在“予力全球每一人、每一组织成就不凡”的使命驱动下,微软正在为实现人工智能与人类智能共进化而不懈努力。洪小文强调,人类智能和人工智能之间的作用有着明显的区别,这意味着,人类是利用我们的共鸣、创造和决定能力来推动未来共同进步的关键。

7月7日,全球人工智能和机器人峰会在深圳举办,大会共聚集了来自全球30多位AI领域的科学家及近300家AI明星企业。微软亚洲研究院资深研究员梅涛在会上发表了主题演讲。在演讲中,他分享了和大家一起探讨“视频内容”这个话题的原因,并从视频内容的生命周期,即视频的创作、处理和消费三方面分别讲述了各自的研究难点以及微软亚洲研究院的解决方案及研究成果。

7 月 17 日,微软亚洲研究院的一篇论文入选计算机视觉领域顶级会议 ICCV 2017。论文中提出了一种全新的通用卷积神经网络交错组卷积模块(IGC),解决了神经网络基本卷积单元中的冗余问题,可以在无损性能的前提下,缩减模型、提升计算速度,有助于深度网络在移动端的部署。微软亚洲研究院视觉计算组主管研究员王井东向机器之心详细解读了这篇论文中的研究工作。

微软亚洲研究院网络图形组培养了一大批中国图形学的人才,近日,其首席研究员童欣接受了新智元的专访。他从多方面展现了计算机图形学的研究和发展情况,结合深度学习技术观测了目前的领域格局,并分析了计算机图形学在具体方向的应用情况。在微软做了近20年图形学研究的童欣表示,目前,网络图形学已实现基础层次研究,未来发展态势良好,但仍需国内图形学界的共同努力。

3D模型聚合网Sketchfab最近就微软亚洲研究院网络图形组与高校一起研发的BendSketch,与微软亚洲研究院主管研究员刘洋和副研究员潘浩进行了电话采访。BendSketch可以用绘制二维线的方法来定义空间中的三维物体。其过程是:首先先制定一条决定形式外边界的关键线,然后通过其他一系列的线,主要包括表面弯曲的笔画、隆起线和尖锐特征来定义曲率。用户可以在几秒内就画出复杂的三维形状。

人工智能已经成为全球科技发展的下一个变革“奇点”,而AI技术人才则是主导这一变革的核心力量。近日,LinkedIn发布了业内首份《全球AI领域人才报告》,基于领英全球5亿高端人才大数据,对全球AI领域核心技术人才的现状、流动趋势和供需情况做了一系列深入分析。微软亚洲研究院首席研究员林钦佑出席了报告发布会,与业界人士共同探讨了中国人工智能领域人才培养、中美在AI领域的竞合以及AI技术趋势等热点话题。

Media Coverage

ZDNet Korea conducted a written interview with Hsiao-Wuen Hon, Corporate Vice President of Microsoft and Managing Director of MSR Asia. The article thoroughly covered Microsoft’s ever-growing engagement with pursuing the coevolution of humans and AI based on the company’s philosophy of empowering every person and organization to achieve more. Dr. Hon highlighted that there are clear distinctions between the roles of humans and AI, and that humans are the key to driving how the coevolution will unfold in the future by utilizing our abilities to resonate, create and determine.

On July 7th, CCF-GAIR 2017 was held in Shenzhen as scheduled, and was attended by more than 30 scientists and nearly 300 star companies from the AI field. Tao Mei, senior researcher at MSR Asia, delivered a keynote speech on the topic of video content. He first explained the reason why he chose this topic, then described the main difficulties encountered during the research process by looking at the life cycle of video contents: creation, processing and consumption. He also gave a thorough introduction to MSR Asia’s solutions to these difficulties and the subsequent research results obtained.

Synced invited Jingdong Wang, lead researcher at MSR Asia, to illustrate his team’s work on the paper selected for ICCV 2017. The paper presented a new universal Convolutional Neural Network module, named Interleaved Group Convolutions (IGC), which solves the redundancy problem in the basic convolution unit of the neural network and more efficiently reduces the model, improving the speed of computing. IGC will help with the deployment of the Deep Neural Network on the mobile side.

AI Era recently conducted an interview with Xin Tong, principal researcher at MSR Asia. He introduced the research and development of computer graphics, and observed the current the field structure combining with deep learning, and analyzed the applications of computer graphics in specific directions. Xin said that currently, internet graphics has realized a basic level of research, and future development is looking promising, but the joint efforts of the domestic graphics industry is needed.

Sketchfab, a leading publishing platform for 3D scenes, conducted a phone interview with lead researcher Yang Liu and associate researcher Hao Pan to learn about BendSketch, a technology developed by MSR Asia and university partners. BendSketch is a two-dimensional line drawing approach to define three-dimensional objects in space. The process is as follows: lay down key lines that define the outer boundary of a form, and then use a series of other lines, including mostly surface bending strokes, ridges/valleys and sharp features, to define curvature. Users can thus “draw” complex three-dimensional shapes in seconds.

AI has become the next "singularity" of global technological development, and AI technical talents will be the core force to dominate this change. Recently, LinkedIn released the first Global AI Talent Report, which was created based on its 500-million global, high-end talents database. The report included in-depth analyses of current trends, and the supply and demand situation of global AI talents. Chin-Yew Lin, principal researcher at MSR Asia, attended the report release conference and explored the personnel training situation, Sino-US cooperation, technological trends, and other hot topics in the AI field.

“博”闻创新

  • 微软沈向洋:以人工智能服务人类社会,微软责无旁贷

    在微软于英国伦敦举办的人工智能发布会上,微软全球执行副总裁沈向洋发表了主题演讲。他表示基于人工智能驱动的技术更加聪明,企业需要以更负责任的方式迎接技术进步,并由此宣布新成立微软研究院人工智能中心。这支由科学家和工程师组成的团队将在一起迎战技术难题的同时致力于加速将最新研究成果转化为产品与服务,让技术尽快造福社会。

  • 基于递归注意力模型的卷积神经网络:让精细化物体分类成为现实

    精细化物体分类存在巨大需求和广阔的应用前景,而目前精细化分类的方法各有局限。微软亚洲研究院多媒体搜索与挖掘组的研究员们提出了“将判别力区域的定位和精细化特征的学习联合进行优化”的构想,搭建了基于递归注意力模型的卷积神经网络结构,它可以更精准地找到图像中有判别力的子区域,采用高分辨率、精细化特征进行描述,提高物体分类的精度。

  • 程序修复程序:78.3%的“Bug修复正确率”是怎么做到的?

    在程序编写中,找Bug、改Bug其实比写程序本身要麻烦的多。为此,北京大学、微软亚洲研究院和电子科技大学的研究人员联合开发了一种新技术ACS (Accurate Condition Synthesis)。该技术可以全自动修复软件系统中的缺陷,无需用户的干预。在缺陷修复基准数据集Defects4J上,ACS修复正确率达到了78.3%。

  • ACL 2017论文研讨会:聚焦自然语言处理领域的新技术

    7月21日,微软亚洲研究院举办了“ACL 2017论文研讨会”,旨在促进自然语言处理相关研究者之间的交流,探讨自然语言处理领域的新技术。微软亚洲研究院已被ACL 2017大会录用论文的作者在研讨会上分别就各自的论文内容进行了分享,并就分享论文的问题、算法、实验等方面进行了深入的交流和讨论。微软亚洲研究院自然语言计算组的研究员通过本文对研讨会内容进行了分享。

  • 走近最前沿的AI科研项目——北京篇 / 台湾篇

    近期,2017 微软亚洲研究院“合作创·研”年会相继在北京微软大厦和台湾宜兰举行,该活动旨在为学者和微软亚洲研究院的研究员们提供展示、交流、探讨的平台,深化项目合作,交流研究经验,分享研究成果,共同探讨亚太学术界共同关注的、最前沿的、具有社会意义的研究课题,以提升亚太学术界在整个计算机领域的影响力。

AI Headlines

  • Microsoft’s role at the intersection of AI, people and society

    At the AI event in the U.K., Harry Shum, Microsoft EVP, said that as the technology employing AI gets smarter, enterprises should ensure that they take a responsible approach to progress. He then announced the establishment of Microsoft Research AI. The team of scientists and engineers at the center will work closely with colleagues across Microsoft’s research labs and product groups. The team will tackle some of the hardest problems in AI and accelerate the integration of the latest AI advances into products and services that benefit customers and society.

  • Recurrent Attention Convolutional Neural Network: Making fine-grained image recognition a reality

    There are great demands and promising prospects for fine-grained image recognition. However, present methods all have their limits. Researchers from MSR Asia’s Multimedia Search and Mining Group proposed a novel recurrent attention convolutional neural network, which recursively learns discriminative region attention and region-based feature presentation in multiple scales. Extensive experiments have demonstrated its superior performance in attention localization and fine-grained recognition tasks.

  • Debugging program: How to achieve a bug repair rate of 78.3% ?

    In programing, debugging is more bothersome than coding. Therefore, researchers from MSR Asia, Peking University and the University of Electronic Science and Technology of China jointly developed a new technology – ACS (Accurate Condition Synthesis), which can automatically repair bugs in software systems without user intervention. Using this approach, they successfully repaired 18 defects in four Defects4J projects. More importantly, the accuracy of this approach reaches 78.3%.

  • ACL 2017 Paper Sharing Workshop: Focusing on new technologies in natural language processing

    ACL 2017, the top international conference in natural language processing, was held in Vancouver, Canada from July 30th to August 4th. On the eve of the conference, MSR Asia conducted the ACL 2017 Paper Sharing Workshop to enhance communications between researchers in the natural language processing field, and to explore new technologies. Researchers and interns from MSR Asia, whose paper was selected for ACL 2017, shared their algorisms and experiments covered by the paper in depth.

  • Approaches to the most advanced AI research projects – Beijing chapter/ Taiwan chapter

    Recently, MSR Asia Symposium on Collaborative Research was held in Beijing and Taiwan Ilan in succession. The event aims to provide a platform for scholars and researchers from MSR Asia in various forms such as theme reporting, round table discussion, project demo and sub forums, to deepen cooperation, exchange research experience and explore the most advanced research subjects with common concern and social significance jointly, enhancing the influence of the Asia Pacific academia in computer field.

“微”言梦想

混合现实遇上深度学习

很多智能设备都受到处理器限制而影响使用,日前,微软全球执行副总裁沈向洋博士在2017 CVPR大会中表示,微软下一代HoloLens将配置HPU 2.0(全息处理单元),包含一个AI协处理器,能够在设备端实现深度神经网络。该处理器能够持续运作,为用户提供更流畅的体验。

为安全而生:用人工智能检测软件安全风险

面对系统错误、崩溃和恶意攻击等一系列麻烦,保护系统免受攻击成为软件开发者们关注的重点。而日益增加的需要测试的软件数量让安全专家们越来越力不从心,为此,微软正在开发一种云服务,开发者可以利用它既快又准地识别出即将发布或已投入使用的软件中的错误和其它安全漏洞。

雷鸣:人工智能时代,科研如何创造产业价值

近年来,关于AI的讨论和研究愈加火热,在技术变更迭起的社会环境中,如何平衡学术研究和产业运作成为难题。近日,“百度七人创始团队之一”的雷鸣应邀来到微软亚洲研究院,分享了他在人工智能领域积累的经验和观点,并在科研和产业的差异分析中总结出了给科研工作者们的几点建议。

CMU计算机学院院长Andrew Moore带你回顾与展望2017年的AI发展

近日,卡内基梅隆大学计算机学院院长Andrew W. Moore和副院长Philip L. Lehman来到微软亚洲研究院,分享了过去两年中人工智能领域的一些技术突破,并从业界和学界两个角度,探讨了他们对于人工智能接下来发展方向的看法。

Eric Horvitz:AI前进的道路原本就应以理智的好奇、审慎与合作为指引

AI已经逐渐深入到了人类日常生活中,在为生活带来便利的同时,也引发了不少人的恐慌。面对AI的发展,我们到底应该拥抱还是担忧?微软研究院院长及技术院士Eric Horvitz发表文章《AI,人类和社会》,为大家解读AI前进的道路。

编译器错误可能引发安全漏洞

面对数量庞杂的代码和文件,程序猿通常会忽视编译器的安全问题,然而编译器可能因不堪代码重负而导致严重的安全漏洞,大规模入侵应用程序。微软研究人员正在开发一款能够自动验证优化以及编译器输出是否正确的软件,不仅支持验证老代码,同时能够提供额外安全保障。

微软开源嵌入式机器学习库ELL:把人工智能扩展到边缘设备

微软美国雷德蒙和印度班加罗尔的研究团队开源了一套嵌入式学习库ELL,帮助开发者把深度神经网络部署到边缘设备上,以消除网络延迟的影响。微软希望向日常用户提供强大的机器学习工具,实现AI的普及和全民化。

微软在ACM EC’17

6月26日-30日,第18届ACM计算经济学会议在MIT举办。被接受的论文中,有16篇来自微软,内容包括了反映人们如何对市场变化做出反应的市场设计模型;双重拍卖投注机制的提出;以及在供不应求的情况下,如何分配保障性住房的理论模型等问题。

Weibo Highlights

When mixed reality meets deep learning

Many intelligent devices are limited by their processors. At 2017 CVPR, Harry Shum, Executive Vice President of Microsoft, announced HPU 2.0, which includes an AI Coprocessor to implement DNNs on future versions of HoloLens. This can provide users with continuous operation and smoother user experience.

AI for security: Microsoft Security Risk Detection makes debut

Protecting systems from attacks is a focus of software developers. Microsoft is creating a cloud service that developers can use to look for bugs and other security vulnerabilities in the software they are preparing to release or use.

Lei Ming: How can research create industry value in the AI era?

Achieving a balance between academic research and industrial operation is a difficult problem. Lei Ming, one of Baidu’s founders, was invited to MSR Asia to share his views and provide advice to researchers based on his experience.

Andrew Moore, Dean of CMU’s School of Computer Science, reviews and forecasts AI development in 2017

Recently, Andrew Moore, Dean of CMU’s School of Computer Science, and Philip L. Lehman, Associate Dean, visited MSR Asia. They reviewed various breakthroughs made in the past two years and proposed their viewpoints on future development in the AI field.

Eric Horvitz: The path forward for AI should be guided by intellectual curiosity, care, and collaboration

As AI gradually penetrates into our daily life, excitement for AI has been tempered by concerns. Whether it deserves our embrace or concern remains a question. Eric Horvitz, Managing Director of MSR AI, published an article titled “AI, People, and Society” to interpret this issue.

A compiler error may cause a security vulnerability

Faced with numerous codes, programmers often ignore compiler security issues, but compilers can break down and cause serious security vulnerabilities. Microsoft’s researchers are developing a software that can automatically verify, optimize and determine whether compiler outputs are correct, not only to validate old codes, but also to provide additional insurance.

Microsoft’s embedded machine learning library ELL: Extending AI to edge devices

A research team from Microsoft Redmond and Bangalore, India is developing an embedded machine learning library, ELL, to help developers deploy the deep neural network on edge devices. Microsoft hopes to provide powerful machine-learning tools to everyday people to realize the democratization of AI.

Microsoft at ACM EC’17

The 18th ACM EC was held from July 26th to 30th at MIT. Microsoft scientists co-authored 21% of the accepted papers, or 16 out of 75. Contents include a market design model that reflects how people react to market changes; a double clinching auction betting mechanism; and a theory model for allocating indemnificatory housing under a shortage, etc.

欢迎订阅我们

输入邮箱地址,即可每月收到期刊更新哦!

订 阅