Agenda with Abstracts - Microsoft Research

Agenda with Abstracts

Microsoft Research Faculty Summit 2013

Monday, July 15, 2013

Time	Event/Topic	Location
8:00–9:00	Continental breakfast
9:00–10:30	Opening Plenary Session	Kodiak
9:00–9:15	Welcome and Introduction
9:15–10:30	Keynote
10:30–10:45	Break
10:45–Noon	Breakout Sessions
	Prediction Engines Chair: David Pennock, Microsoft Research Speakers: Sanmay Das, Virginia Polytechnic Institute & State University; Robin Hanson, George Mason University; and Nicolas Lambert, Stanford University From crowdsourcing to data-driven models, technology offers new ways to collect and aggregate information on an unprecedented scale, allowing researchers to make reliable predictions about elections, policy, corporate decisions, economics, finance, sports, and entertainment. This session will report on progress in the science and engineering of polls, prediction markets, and forecast models. Examples include a fundamental model that correctly predicted 49 of 50 states nine months before the US Presidential election, and combinatorial prediction markets capable of estimating billions of complex predictions like correlations between states. The session will include talks from academic experts in the field as well as Microsoft researchers investigating prediction markets, scoring rules, polls, and forecast models.	Cascade
	Efficiently Verifying Outsourced Cloud Computation: from Theory to Practice Chair: Bryan Parno, Microsoft Research Speakers: Michael Mitzenmacher, Harvard University; Michael Walfish, University of Texas at Austin The growth of cloud computing contributes to a desire to outsource computing from a client device to an online service. However, the client should be able to efficiently verify the correctness of the results returned, to guard against malicious or malfunctioning services. The session speakers have each built a different system for verifying arbitrary outsourced computations. Each draws on a different strand of theory and applies different cryptographic and system-engineering techniques. Collectively, their efforts have brought the cost of verification down by over 20 orders of magnitude in the last three years, making verifiable computation close to practical for a variety of applications.	Rainier
	Database Systems Exploiting New Hardware Platforms Chair: David Lomet, Microsoft Research Speakers: Paul Larson, Microsoft Research; Justin Levandoski, Microsoft Research; Thomas Neumann, Technische Universität München; and Jignesh Patel, University of Wisconsin, Madison Main memory databases have become increasingly important as a way to provide high-performance OLTP (on-line transaction processing) capability. By restricting data to by main memory resident, rather than storing disk resident data entirely in the main memory cache, it is possible to use new technology that results in much better performance. This new setting calls for a revolution in the access methods, concurrency control, and recovery. It reduces the need for high scalability because peak performance is so much higher. This session will explore main memory technologies, the performance that they produce, the problems encountered, and the prospects going forward.	St. Helens
	Data in Disasters—the Potential and the Limits Chair: Kate Crawford, Microsoft Research Speakers: Eytan Adar, University of Michigan-Ann Arbor; Andres Monroy-Hernandez, Microsoft Research; and Leysia Palen, University of Colorado-Boulder There has been a rapid expansion in the ability to gather and analyze communication data during acute events. As we saw most recently during Hurricane Sandy, people use social media services like Twitter and Facebook to share a range of personal information: be it their location, their health, the status of power and water access, or images of the disaster in their neighborhood. Fields such as crisis informatics and machine learning have brought powerful new insights to how communities deal with crises by drawing on this data. However, there are also causes for concern, such as the privacy and longevity of that data, and the invisibility of the many people and communities that are either not using social media or are living in areas with damaged or overwhelmed ICT infrastructure. This panel draws together a diverse set of individuals who study information use and sharing during acute events. Our panelists will explore research avenues for disaster data, including the difficulties with the reliability of citizen accounts of disaster, issues with privacy and information sharing during disasters, and the hidden biases of disaster data and what it cannot capture.	Baker
Noon–1:00	Lunch and Engaging with Microsoft Research
	Engaging with Microsoft Research	Hood
	Lunch
1:00–2:30	Breakout Sessions
	Interaction for Machine Learning Chair: Ashish Kapoor, Microsoft Research Co-organizers: Ece Kamar, Dan Bohus, Lihong Li, and Jason Williams Speakers: James Fogarty, University of Washington; Lise Getoor, University of Maryland; Patrice Simard, Microsoft Research; and Jerry Zhu, University of Wisconsin, Madison The traditional role of the human operator in machine learning problems is that of a batch labeler, whose work is done before the learning even begins. However, humans can provide guidance to a learning system via a richer set of inputs and interact directly with the learning algorithm as it learns. Active research problems in this space include designing interactions for acquiring human guidance, active and life-long learning, interactive clustering, query by selection, learning to rank, and effective crowdsourcing. Addressing these problems requires contributions from multiple disciplines such as human-computer interaction, artificial intelligence, and machine learning. In this session, we bring together researchers from different disciplines and identify how richer forms of interaction with humans could help machine learning systems, in terms of both surveying the major paradigms and sharing information about new work in this area. Through a combination of presentations and discussions, we hope to gain a better understanding of the available algorithms and best practices, of their inherent limitations, and of challenges and opportunities for future research in this space.	Cascade
	The Economics of Computing Chair: Nicole Immorlica, Microsoft Research Speakers: Dan Huttenlocher, Cornell University; Muthu Muthukrishnan, Microsoft Research; and Eva Tardos, Cornell University Many computational tasks require the participation of a diverse set of users. These users are motivated by their own interests and act to maximize the utility they gain from interacting with the system. Examples abound, including online advertising auctions, financial transactions, voting platforms, user-generated content sites, and cloud-computing systems. This session focuses on how to design such systems with appropriate economic incentives for users.	Rainier
	Hosting Blazing Fast Services: From 1 Core to 1M Cores Chair: Victor Bahl, Microsoft Research This session will focus on the systems and networking infrastructure that enable Microsoft to host thousands of services that serve hundreds of millions of users globally with an unprecedented level of performance. Microsoft builds and operates best-in-class data centers that contain 2,500 to more than 100,000 servers (per data center), which are connected by using cutting-edge networks within and across data centers. To meet the performance, cost, and availability demands of Microsoft and third-party services that are hosted in our data centers, our systems and networks have evolved to be highly programmable and software-defined, thereby using both hardware and software to solve challenges. Presenters from Microsoft products groups and Microsoft Research will describe the networking technologies that fuel our data centers. They will share and discuss results of their investigations and deployments, all of which are a part of Microsoft’s worldwide, multi-year infrastructure investment and development quest.	St. Helens
	Computer-Aided Education Chair: Donald Brinkman, Microsoft Research Speakers: Sumit Gulwani, Microsoft Research; Armando Solar Lezama, Massachusetts Institute of Technology; and Zoran Popovic, University of Washington New forms of education are emerging that offer the potential to amplify the reach of a single educator to embrace thousands of simultaneous learners. Organizations like Khan Academy and Coursera are exploring the opportunities and challenges offered by massively open online courseware. In a classroom of thousands, tasks like assignment grading, problem generation, and student analytics can become intractable. In this session, we will explore approaches that seek to mitigate these challenges by applying machine learning, cloud computing, and other innovative technologies.	Baker
2:30–2:45	Break
2:45–4:15	Breakout Sessions
	Machine Learning for Interactive Systems: Practical Challenges and Opportunities Chair: Lihong Li, Microsoft Research Co-organizers: Dan Bohus, Ece Kamar, Ashish Kapoor, and Jason Williams Speakers: Léon Bottou, Microsoft Research, Michael Littman, Brown University; David Traum, University of Southern California; and Manuela Veloso, Carnegie Mellon University In many settings in which agents interact with people, such as conversational systems and software assistants, machine learning problems go beyond mapping inputs to outputs with supervised learning. Substantial progress has been made in areas like artificial intelligence, operations research, and statistics. Yet due to the complex nature of these problems, applications are often hindered by challenges that have not received sufficient attention in the research community. As a result, there have been relatively few high-profile deployments of these promising technologies. The goal of this session is to identify critical bottlenecks for applying machine learning in interactive systems, and to discuss opportunities for future research, informed by case studies of deployed applications. Example topics include challenges that are associated with offline evaluation and optimization, the design of states and reward, and life-long and open-world learning.	Cascade
	Towards a Research Platform for Internet of Things (IoT) Chair: Arjmand Samuel, Microsoft Research Speakers: Arjmand Samuel, Microsoft Research, and Kamin Whitehouse, University of Virginia We are living in a world of connected devices. These devices can take a number of shapes and forms: from tablet computers and smartphones, to sensors in the home, to embedded and wearable computers. Increasingly, there is interest in deploying devices and sensors in homes, buildings, vehicles, and any other spaces where human activity occurs. Most of these devices require software architectures and cloud services that are generally developed for either a specific device or class of devices. The prototyping and development of new devices requires new architectures and services, which consume precious research resources. To alleviate this barrier in research, a flexible platform with expandable device interface as well as programmatic access to device data and functionality, are required. In this session, researchers from academia and Microsoft Research will talk about research challenges in connected devices in the home and beyond, along with an overview of the upcoming HomeOS platform.	Rainier
	Big Data Platforms Chair: Surajit Chaudhuri, Microsoft Research	St. Helens
	Publishing and Perishing in the Twenty-First Century Chair: Alex Wade, Microsoft Research Speakers: Jennifer Lin, Public Library of Science; Filippo Menczer, Indiana University; Jevin West, University of Washington The creation of quantifiable measures of the impact and relative importance of a publication, a journal, an individual researcher’s output, or a university is close kin to ranking algorithms in information retrieval. Eugene Garfield developed the journal impact factor (IF) a half-century ago based on a two-year window of citations. And more recently, Jorge Hirsch invented the h-index to quantify an individual’s productivity based on the distribution of citations over one’s publications. There are also several competing “world university ranking” systems in wide circulation. Most traditional bibliometrics seek to build upon the citation structure of scholarship in the same manner that PageRank uses the link structure of the web as a signal of importance, but new approaches (or Alt Metrics) are now seeking to harness usage patterns and social media to assess impact.	Baker
4:15–4:30	Break
4:30–5:30	Plenary Session	Kodiak
	Keynote	Kodiak
5:30–6:00	Travel to Kirkland Dock
6:00–9:00	Argosy Cruise on Lake Washington

Tuesday, July 16, 2013

Time	Event/Topic	Location
8:00–9:00	Continental breakfast
9:00–10:30	Plenary Session	Kodiak
9:00–9:15	Welcome
9:15–10:30	Keynote
10:30–10:45	Break
10:45–1:00	DemoFest (continues through lunch)	McKinley
Noon–1:00	Lunch and Breakout Session
	Innovations and Research Highlights in Attracting Women in Computing Chair: Rane Johnson, Microsoft Research Speakers: Tiffany Barnes, North Carolina State University; Lucy Sanders, National Center for Women Information Technology; and Constance Steinkuehler, University Wisconsin at Madison With 1.4 million open jobs in computing by 2018 and only 29 percent of those expected to be filled by women, there has been much research and investigation into how to increase the pipeline. Join top researchers and experts in the field who are making strides in this critical area and discuss the challenges and the research collaborations with Microsoft Research that are making a difference in growing more women in computing. Participants will have the opportunity to become active in the movement to facilitate greater diversity among computer scientists. In addition, participants will be invited to view two DemoFest booths to learn from women rising stars who are making a difference in the world through computing research and solutions.	Cascade
	Lunch
1:00–3:30	Design Expo Chair: Lili Cheng, Microsoft Research	Kodiak
	Making Data Useful: Improving Your Life, Community, and World We live in a world that is increasingly alive with sensors and data. The big data, sensor networks and transparency movements have left us with a glut of potentially useful free data that is lying fallow. How can we use this to improve life, local community, and the world at large? People and devices collect and share data both passively—monitoring environmental change—and actively—explicitly capturing and sharing information, combining data with other information, and using in unexpected ways. Today, we are only starting to understand how best to put this data to work to improve our lives and the world around us. How might data—particularly information that makes civic society run, like bus schedules, election cycles, political information, first-hand reporting, volunteer logistics, and sporting and media events—make for a better and more community-oriented place to live? How might this data help us understand the past and present and predict the future with more precision, and how might an individual’s personal data be used to help filter and make information more relevant in different contexts or situations? What are key problems this data can be used to help solve, what new troubles can we anticipate it creates? Carnegie Mellon University, School of Design, Pittsburg, PA, United States Professors: Peter Scupelli, Bruce Hanington Interdisciplinary Center (IDC), Herzliya, Israel Professor: Oren Zuckerman National Institute of Design (NID), Ahmedabad, India Professor: Bibhudutta Baral New York University, Interactive Telecommunications Program, New York, NY, United States Professor: Clay Shirky Northumbria, Newcastle upon Tyne, United Kingdom Professor: Trevor Duncan Technische Universiteit (TU) Eindhoven, Eindhoven, The Netherlands Professor: Berry Eggen Universidad Iberoamericana, Design Department, Mexico City, Mexico Professor: Jorge Meza Aguilar University of California at Los Angeles, Design Media Arts, Los Angeles, CA, United States Professor: Christian Moeller University of Washington, Interactive Design Division, Seattle, WA, United States Professor: Axel Roesler	Kodiak
1:00–2:30	Breakout Sessions
	The Coming Genomics Software Revolution? Chair: Bill Bolosky, Microsoft Research Speakers: David Haussler, University of California, Santa Cruz; David Heckerman, Microsoft Research; David Patterson, University of California, Berkeley; and George Varghese, Microsoft Research With the cost to sequence a full human genome soon to fall below US$1,000, most people will have their DNA sequenced and stored in a database along with their medical records. The hardware revolution has largely occurred; this session will focus on the software issues that remain. If there were a database of a million or more fully sequenced genomes together with phenotype information such as disease and treatment codes, what software would be necessary to mine it for biological insights that might eventually lead to drug discovery and personalized medicine? Cancer is a fundamentally genomic disease, and drugs such as Gleevec have shown the effectiveness of combatting the genomic pathways of disease. Could a large genomic database enable progress like this for other cancers, or for diseases other than cancer? What are the medical and systems implications of such a database? Could genomics engender a new software industry? We will discuss the existing TCGA (The Cancer Genome Atlas) database currently consisting of 5,000 cancer genomes and the open algorithmic problems that arise in making use of the data. We will describe parallel infrastructure and computing paradigms for genome computation, and new systems issues that must be solved. We will discuss issues in scaling genomic inference in the face of confounding factors. We will also describe challenges that must be solved in order to query a vast genome database interactively in a few seconds.	Cascade
	Modern Programming for and via the Web Browser Chair: Judith Bishop, Microsoft Research Speakers: Shriram Krishnamurthi, Brown University; Steve Lucco, Microsoft; Tom Ball, Microsoft Research Programming for web browser is a complex affair, in the same way that programming of early computers via assembly language was before the advent of higher-level languages such as FORTRAN and C. Today, languages such JavaScript, HTML, and CSS form the basic low-level programming abstractions. This session describes research and technology that makes programming for the web easier and safer. We focus on new developments in JavaScript, investigate the latest changes in TouchDevelop for mobile touch devices, and announce TypeScript, a new language for application-scale JavaScript development.	Rainier
	From Data Science to Data Intelligence Chair: Evelyne Viegas, Microsoft Research Speakers: Léon Bottou, Microsoft Research, and Percy Liang, Stanford University Today, we live in a data-driven world and new directions in data-driven research have already revolutionized big data applications such as gaming, Internet vision, machine translation, and spell checking by bringing machine learning to the core of the information revolution. But to continue to drive innovation, three fundamental aspects of data-driven research need to change: data must be made available to the research community to allow for benchmarking, reproducibility and transparency of experimentation need to happen to enable science, and researchers need to work on solving big and real world problems to advance the state of the art. This session addresses all three aspects of data-driven research by exploring the status of data sharing in data science, and looking at what data intelligence may be.	St. Helens
	Visual Recognition Chair: Andrew Blake, Microsoft Research Speakers: Kristen Grauman, University of Texas at Austin; Fei-Fei Li, Stanford University; and Larry Zitnick, Microsoft Research The fields of Computer Vision and Machine Learning are becoming increasingly intertwined, with many of the recent breakthroughs in object and scene recognition coming from the availability of large labeled datasets and sophisticated machine learning techniques. In this session, leading researchers in these fields share their perspectives on recent advances and current challenges. How do sophisticated machine learning approaches aid in solving difficult recognition problems? What role do large labeled datasets and recognition challenges play in advancing the state of the art and enabling data-driven approaches to recognition? And how can the layout of objects in a scene as well as relationship to natural language models give us an edge in describing complex scenes with multiple actors and objects? These are just some of the questions at the forefront of this rapidly evolving research field.	Baker
2:30–2:45	Break
2:45–4:15	Breakout Sessions
	Deep Machine Learning: a Panel Chairs: Li Deng, Microsoft Research, and John Platt, Microsoft Research Speakers: Yoshua Bengio, Université de Montréal; Honglak Lee, University of Michigan – Ann Arbor; Andrew Ng, Stanford University; and Ruslan Salakhutdinov, University of Toronto Deep learning is a sub-field of machine learning that focuses on hierarchical representations of features or concepts, where high-level semantic-like features can emerge via automatic layer-by-layer learning from low-level features. In recent years, deep learning has achieved important successes in a variety of applied artificial intelligence tasks including speech recognition, computer vision, and natural language processing. The implications of such recent work have been prominently covered in recent media with both enthusiasm and skepticism. Since 2009, in partnership with academics, Microsoft Research has been pursuing deep learning research and technology transfer, and has pioneered the development of industry-scale deep learning technology for speech recognition. It is useful to share our experience with wider academic communities and learn from each other. To make the material and directions of interest to a broader computer science audience, we plan to offer a tutorial to demystify the “black-art” label often attached to deep learning.	Cascade
	Spam Wars: the Systems Strike Back Chair: Judith Bishop, Microsoft Research Speakers: Saikat Guha, Microsoft Research; Ben Livshits, Microsoft Research; and Stefan Savage, University of California, San Diego How do we safeguard user privacy in a way that does not disrupt the emerging business models behind cloud computing, online social networks, and mobile ecosystems? Numerous efforts focus on identifying and blocking individual abusive advertising mechanisms, but there is a parallel research direction that undermines the associated means of monetization: payment networks. This session brings together the latest attempts to pit software against software in the malware and spam wars. How do we prevent vulnerabilities in modern cars with electronic keys, or electronic tire- or brake- sensors? We’ll discuss how to combine the goals of privacy and content personalization, especially on mobile devices.	Rainier
	Quantum Computing: the Next Frontier Chair: Krysta Svore, Microsoft Research Speakers: Scott Aaronson, Massachusetts Institute of Technology; Charles Marcus, Niels Bor Institute; and Matthias Troyer, Eidgenössische Technische Hochschule Zürich In 1981, Richard Feynman proposed a device called a “quantum computer” that would take advantage of methods founded on the laws of quantum physics and promise computational speed-ups over classical methods. In the last three decades, quantum algorithms have been developed that offer fast solutions to problems in a variety of fields including number theory, optimization, database search, chemistry, and physics. For quantum devices, this past year marks a significant breakthrough. Recent experiments point to the observation of an elusive particle at the heart of several scalable device proposals, called the Majorana fermion. Advancements in understanding the role of quantum speed-ups in the commercial D-Wave One quantum processor, which has been the subject of intense debate, have also been made. This session will showcase quantum algorithms with real-world applications and highlight breakthroughs in quantum devices, including Majorana-based devices and the D-Wave quantum processor.	St. Helens
	Visual Motion and Structure Chair: Rick Szeliski, Microsoft Research Speakers: Bill Freeman, Massachusetts Institute of Technology; Shahram Izadi, Microsoft Research; and Noah Snavely, Cornell University For several decades, the analysis of visual motion and 3-D scene structure have been central to the study of computer vision. Recent breakthroughs in analyzing and amplifying subtle motions can now give us insights into previously unstudied phenomena such as natural (visual) biometric signals and basic structure physics. Similarly, the ability to create detailed large-scale models either from photo collections found on the web or from cheap depth cameras (Kinect) opens up new applications in visualization and the exploration of 3-D locations. The session speakers will focus on recent technical breakthroughs, promising applications, and remaining challenges and opportunities in these areas.	Baker
4:15–4:30	Break
4:30–5:30	Closing Plenary Session	Kodiak
	Keynote	Kodiak
5:30–7:00	Networking Reception

Top of page