![\"CNAM](\"http:\/\/project.inria.fr\/siliconvalley\/files\/2014\/03\/CNAM_\u00a9-Inria-Photo-G.-Cohen.jpg\")
<\/a>CNAM \u00a9Inria Photo G. Cohen<\/p><\/div>\n
Berkeley \u2013 Inria \u2013 Stanford Workshop 2014<\/strong><\/h1>\n\n- Date<\/strong>: June 17-19, 2014<\/li>\n
- Location: <\/strong>CNAM, Paris<\/li>\n<\/ul>\n
The fourth workshop Berkeley-Inria-Stanford \u00a0(BIS\u20192014) is part of the joint research program Inria@SiliconValley. It is co-organized by Inria, Stanford University, and UC Berkeley in partnership with CITRIS and the French Ministry of Foreign Affairs. \u00a0The objectives of the workshop are to present the current status of ongoing scientific collaborations and to work on proposals for future ambitious joint projects. Further detail about the workshop may be obtained from its organizing committee<\/a>.<\/em><\/p>\nProgram<\/strong> :<\/strong><\/em><\/span><\/h2>\nDOWNLOAD FULL PROGRAM<\/strong><\/a><\/p>\n\n- \n
June 17th<\/strong> (session open to public):\u00a0Opening keynotes highlighting results of ongoing collaborations<\/h4>\n<\/li>\n- 13:30: Registration<\/strong><\/li>\n
- 14:00-14:20: Formal Opening: Michel Cosnard, CEO, Inria and H\u00e9l\u00e8ne Kirchner, International relations Director, Inria<\/strong><\/li>\n
- \n
14:20- 15:05: Analysis of Data-Centric Workflows by Victor Vianu<\/a><\/strong>(UC San Diego, Inria International Chair):<\/p>\n\n- <\/strong>\n
![\"Victor](\"https:\/\/project.inria.fr\/inria-siliconvalley\/files\/2014\/04\/Victor-Vianu-144x150.jpg\")
<\/a>Victor Vianu<\/strong><\/p><\/div>\nWorkflows centered around data have become pervasive in a wide variety of applications, including health-care management, e-commerce, business processes, scientific workflows, and e-government. Such workflows are often very complex and involve numerous interacting actors. They are prone to costly bugs, whence the need for static analysis in order to verify critical properties. Analysis tools are also needed\u00a0 to facilitate the integration, interoperation and evolution of workflows, and to provide runtime assistance to participating actors. \u00a0This talk will present an overview of recent research carried out at UC San Diego and INRIA on the analysis of data-centric workflows, an area of growing interest in both academia and industry. View presentation<\/strong><\/a><\/em><\/li>\n<\/ul>\n\n- 15:05 -15:50: Interactive Machine Learning by<\/strong> John Canny <\/b><\/a>(U.C. Berkeley, Inria International Chair):<\/strong>\n
![\"John](\"https:\/\/project.inria.fr\/inria-siliconvalley\/files\/2014\/04\/John-Canny-150x150.jpg\")
<\/a>John Canny<\/strong><\/p><\/div>\nMachine learning is now an essential tool in business and the sciences. \u00a0And many such applications involve “Big Data” of scales that stretch current technologies. Much of the value from Big Data is untapped, and will require interactive tools that support rapid exploration and hypothesis-testing. The current wave of analytic tools for Big Data rely primarily on cluster computing for acceleration. But they sacrifice responsiveness and often use only a fraction of the hardware’s theoretical performance. The BID Data project has developed new, hardware-accelerated learning tools (BIDMach) with much higher efficiency and real-time performance. Single machine speeds for BIDMach on many common learning tasks exceed that of any reported cluster system. \u00a0Furthermore, we show that these gains can be scaled further on clusters of machines using new, faster communication primitives. \u00a0Interactive analysis of massive datasets is now possible, and has many advantages: exploration of models can be guided at high-level by the analyst; richer more realistic loss functions can be used; and models parameters can be tuned in one pass against a variety of key performance indicators. The design of BIDMach relies on deep co-design of algorithms and hardware. One example is “cooled Gibbs sampling”: a new approach to parameter inference, which is simple, fast and well-matched to modern SIMD (e.g. GPU) hardware. Another is “butterfly mixing”, a fast and highly fault-tolerant approach to synchronizing learning algorithms across a cluster<\/em>. View Presentation<\/strong><\/em><\/a><\/li>\n- 15:50-16:20: Coffee Braek<\/strong><\/li>\n
- 16:20-17:05: ORESTE- Optimization based control of networks of discretized PDEs: Application to traffic engineering by Paola Goatin<\/a> (Inria) & Nikolaos Bekiaris-Liberis<\/a> (U.C. Berkeley):\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 <\/strong><\/strong><\/strong>\n
![\"Paola](\"https:\/\/project.inria.fr\/inria-siliconvalley\/files\/2014\/04\/paola-Goatin-150x150.jpg\")
<\/a>Paola Goatin<\/strong><\/p><\/div>\n![\"Nikolaos](\"https:\/\/project.inria.fr\/inria-siliconvalley\/files\/2014\/04\/Bekiaris-e1398848751100-150x150.jpg\")
<\/a>Nikolaos Bekiaris<\/strong><\/span><\/p><\/div>\nT<\/em><\/em>he talk will review the results obtained by the ORESTE Associated Team during its t<\/em><\/em>hree years life-span. We deal with macroscopic models of traffic flow on road<\/em><\/em>networks descr<\/em><\/em>ib<\/em><\/em>ed by conservation laws\u00a0coupled with suitable<\/em><\/em>b<\/em><\/em>oundary conditions at junctions. In particular, we focus <\/em><\/em>on<\/em><\/em>the development of an<\/em><\/em>efficient coordinated ramp-metering strategy resulting in better traffic performances than the\u00a0state-of-the-art practitioners tools. We also extend the methodology<\/em><\/em>to the case of multi-<\/em><\/em><\/em>commodity\u00a0flows intended to reduce traffic congestion by re-routing compliant users<\/em><\/em>.<\/em><\/em><\/li>\n- 17:05-17:50: CRISP – Leveraging Artistic and Perceptual Insights for<\/b>Image Synthesis by George Drettakis<\/a> & Adrien Bousse<\/a><\/b>au<\/a> (Inria) :<\/strong><\/strong>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0<\/strong><\/strong>\u00a0 <\/strong><\/strong>\n
![\"Adrien](\"https:\/\/project.inria.fr\/inria-siliconvalley\/files\/2014\/04\/Adrien-Bousseau-145x150.jpg\")
<\/a>Adrien Bousseau<\/strong><\/p><\/div>\n![\"George](\"https:\/\/project.inria.fr\/inria-siliconvalley\/files\/2014\/04\/George-Drettakis-PhotoC2013-anna-Drettakis-150x150.jpg\")
<\/a>George Drettakis<\/strong> <\/span>Photo \u00a92013 Anna Drettakis<\/span><\/p><\/div>\n<\/strong><\/strong><\/strong><\/strong><\/strong><\/strong>The goal of the CRISP associate team between REVES and UC Berkeley is to<\/em><\/em><\/em> <\/em><\/em>investigate novel ways to<\/em><\/em>create, render and interact with images bas<\/em><\/em>ed on the study<\/em><\/em>of human perception <\/em>and artistic practice. To achieve this goal<\/em>we focus on understanding how people perceive and draw complex material, lighting and shape, on developing new rendering algorithms based on this understanding, and on building interactive tools that help users create synthetic images. We will illustrate our methodology with four projects developed over the 4 years of CRISP. The first two projects study the perception of materials in stylized rendering and the perception of distortions in image-based rendering. The two other projects facilitate the depiction of materials and transparency effects in vector graphics<\/em>. View presentation<\/strong><\/em><\/a><\/li>\n- 17:50 : Adjourn<\/strong><\/li>\n
- 18:00: Live streaming (to be confirmed): “Computer Science: All questions answered” by Donald Knuth, Stanford University (http:\/\/colloquium.lip6\/knuth-2014-06-07\/<\/a>)<\/strong><\/li>\n<\/ul>\n<\/li>\n<\/ul>\n
\n- \n
June 18th: 9.00am – 5.30pm<\/strong> (For Inria and Inria@SiliconValley partners) : The day is organized around parallel working sessions that will be the opportunity for associated teams to pitch about their key contributions and further exchange about research challenges ahead on the given research themes.<\/h4>\n<\/li>\n<\/ul>\nDOWNLOAD PARALLEL WORKING SESSIONS DETAILED SCHEDULE<\/strong><\/a> The 3 topics under scrutiny are:<\/p>\ni. New computation and data analysis\u00a0<\/strong> <\/span><\/h5>\nChairs<\/strong><\/span>: Jean Roman (Inria) and Henry van den Bendem (Stanford University)<\/p>\n Associate Teams:<\/strong><\/span>\u00a0 AQUARIUS<\/a>, CARDIO, COALA,\u00a0 <\/a>COMFORT<\/a>, FASTLA<\/a>,\u00a0 ITSNAP<\/a>,\u00a0ORESTE<\/a><\/p>\nii. Content <\/strong>and supporting learning technologies<\/strong><\/strong><\/span><\/h5>\nChairs<\/span>:<\/strong> Cordelia Schmid (Inria), John Canny (U.C. Berkeley) –<\/p>\nAssociate Teams:<\/strong><\/span>\u00a0CRISP<\/a>,\u00a0HYPERION,<\/a>\u00a0 OAKSAD<\/a>, SNOWFLAKE, SPLENDID<\/a>,VIP, WILLOW<\/p>\niii. Computation and communication for the future internet <\/strong><\/strong><\/h5>\nChairs<\/strong>:<\/span> Christine Morin (Inria), Deb Agarwal (LBNL)<\/p>\nAssociate Teams<\/span>:<\/strong>\u00a0BIGDATANET<\/a>,\u00a0CLOUDSHARE<\/a>,\u00a0CLOUDY<\/a>,\u00a0COMMUNITY<\/a>, DALHIS<\/a>,\u00a0RIPPES<\/a>, SnFactory, MiMove<\/p>\n\n- \n
June 19th<\/strong> (session open to public) : Toward smarter cities.<\/h4>\n<\/li>\n<\/ul>\n
The fourth workshop Berkeley-Inria-Stanford \u00a0(BIS\u20192014) is part of the joint research program Inria@SiliconValley. It is co-organized by Inria, Stanford University, and UC Berkeley in partnership with CITRIS and the French Ministry of Foreign Affairs. \u00a0The objectives of the workshop are to present the current status of ongoing scientific collaborations and to work on proposals for future ambitious joint projects. Further detail about the workshop may be obtained from its organizing committee<\/a>.<\/em><\/p>\n DOWNLOAD FULL PROGRAM<\/strong><\/a><\/p>\n 14:20- 15:05: Analysis of Data-Centric Workflows by Victor Vianu<\/a><\/strong>(UC San Diego, Inria International Chair):<\/p>\n Victor Vianu<\/strong><\/p><\/div>\n Workflows centered around data have become pervasive in a wide variety of applications, including health-care management, e-commerce, business processes, scientific workflows, and e-government. Such workflows are often very complex and involve numerous interacting actors. They are prone to costly bugs, whence the need for static analysis in order to verify critical properties. Analysis tools are also needed\u00a0 to facilitate the integration, interoperation and evolution of workflows, and to provide runtime assistance to participating actors. \u00a0This talk will present an overview of recent research carried out at UC San Diego and INRIA on the analysis of data-centric workflows, an area of growing interest in both academia and industry. View presentation<\/strong><\/a><\/em><\/li>\n<\/ul>\n John Canny<\/strong><\/p><\/div>\n Machine learning is now an essential tool in business and the sciences. \u00a0And many such applications involve “Big Data” of scales that stretch current technologies. Much of the value from Big Data is untapped, and will require interactive tools that support rapid exploration and hypothesis-testing. The current wave of analytic tools for Big Data rely primarily on cluster computing for acceleration. But they sacrifice responsiveness and often use only a fraction of the hardware’s theoretical performance. The BID Data project has developed new, hardware-accelerated learning tools (BIDMach) with much higher efficiency and real-time performance. Single machine speeds for BIDMach on many common learning tasks exceed that of any reported cluster system. \u00a0Furthermore, we show that these gains can be scaled further on clusters of machines using new, faster communication primitives. \u00a0Interactive analysis of massive datasets is now possible, and has many advantages: exploration of models can be guided at high-level by the analyst; richer more realistic loss functions can be used; and models parameters can be tuned in one pass against a variety of key performance indicators. The design of BIDMach relies on deep co-design of algorithms and hardware. One example is “cooled Gibbs sampling”: a new approach to parameter inference, which is simple, fast and well-matched to modern SIMD (e.g. GPU) hardware. Another is “butterfly mixing”, a fast and highly fault-tolerant approach to synchronizing learning algorithms across a cluster<\/em>. View Presentation<\/strong><\/em><\/a><\/li>\n Paola Goatin<\/strong><\/p><\/div>\n Nikolaos Bekiaris<\/strong><\/span><\/p><\/div>\n T<\/em><\/em>he talk will review the results obtained by the ORESTE Associated Team during its t<\/em><\/em>hree years life-span. We deal with macroscopic models of traffic flow on road<\/em><\/em>networks descr<\/em><\/em>ib<\/em><\/em>ed by conservation laws\u00a0coupled with suitable<\/em><\/em>b<\/em><\/em>oundary conditions at junctions. In particular, we focus <\/em><\/em>on<\/em><\/em>the development of an<\/em><\/em>efficient coordinated ramp-metering strategy resulting in better traffic performances than the\u00a0state-of-the-art practitioners tools. We also extend the methodology<\/em><\/em>to the case of multi-<\/em><\/em><\/em>commodity\u00a0flows intended to reduce traffic congestion by re-routing compliant users<\/em><\/em>.<\/em><\/em><\/li>\n Adrien Bousseau<\/strong><\/p><\/div>\n George Drettakis<\/strong> <\/span>Photo \u00a92013 Anna Drettakis<\/span><\/p><\/div>\n <\/strong><\/strong><\/strong><\/strong><\/strong><\/strong>The goal of the CRISP associate team between REVES and UC Berkeley is to<\/em><\/em><\/em> <\/em><\/em>investigate novel ways to<\/em><\/em>create, render and interact with images bas<\/em><\/em>ed on the study<\/em><\/em>of human perception <\/em>and artistic practice. To achieve this goal<\/em>we focus on understanding how people perceive and draw complex material, lighting and shape, on developing new rendering algorithms based on this understanding, and on building interactive tools that help users create synthetic images. We will illustrate our methodology with four projects developed over the 4 years of CRISP. The first two projects study the perception of materials in stylized rendering and the perception of distortions in image-based rendering. The two other projects facilitate the depiction of materials and transparency effects in vector graphics<\/em>. View presentation<\/strong><\/em><\/a><\/li>\n DOWNLOAD PARALLEL WORKING SESSIONS DETAILED SCHEDULE<\/strong><\/a> The 3 topics under scrutiny are:<\/p>\n Chairs<\/strong><\/span>: Jean Roman (Inria) and Henry van den Bendem (Stanford University)<\/p>\n Associate Teams:<\/strong><\/span>\u00a0 AQUARIUS<\/a>, CARDIO, COALA,\u00a0 <\/a>COMFORT<\/a>, FASTLA<\/a>,\u00a0 ITSNAP<\/a>,\u00a0ORESTE<\/a><\/p>\n Chairs<\/span>:<\/strong> Cordelia Schmid (Inria), John Canny (U.C. Berkeley) –<\/p>\n Associate Teams:<\/strong><\/span>\u00a0CRISP<\/a>,\u00a0HYPERION,<\/a>\u00a0 OAKSAD<\/a>, SNOWFLAKE, SPLENDID<\/a>,VIP, WILLOW<\/p>\n Chairs<\/strong>:<\/span> Christine Morin (Inria), Deb Agarwal (LBNL)<\/p>\n Associate Teams<\/span>:<\/strong>\u00a0BIGDATANET<\/a>,\u00a0CLOUDSHARE<\/a>,\u00a0CLOUDY<\/a>,\u00a0COMMUNITY<\/a>, DALHIS<\/a>,\u00a0RIPPES<\/a>, SnFactory, MiMove<\/p>\nProgram<\/strong> :<\/strong><\/em><\/span><\/h2>\n
\n
June 17th<\/strong> (session open to public):\u00a0Opening keynotes highlighting results of ongoing collaborations<\/h4>\n<\/li>\n
\n
<\/a>\n
<\/a><\/a><\/a><\/a><\/a>\n
June 18th: 9.00am – 5.30pm<\/strong> (For Inria and Inria@SiliconValley partners) : The day is organized around parallel working sessions that will be the opportunity for associated teams to pitch about their key contributions and further exchange about research challenges ahead on the given research themes.<\/h4>\n<\/li>\n<\/ul>\n
i. New computation and data analysis\u00a0<\/strong> <\/span><\/h5>\n
ii. Content <\/strong>and supporting learning technologies<\/strong><\/strong><\/span><\/h5>\n
iii. Computation and communication for the future internet <\/strong><\/strong><\/h5>\n
\n
June 19th<\/strong> (session open to public) : Toward smarter cities.<\/h4>\n<\/li>\n<\/ul>\n