http://research.microsoft.com/apps/dp/vi/videos.aspx Watch the latest lectures from Microsoft Research © 2012 Microsoft Corporation. All rights reserved. en-US Mon, 20 Feb 2012 18:08:25 GMT [Speaker: Souneil Park] In this talk, we introduce our research on a major problem of journalism, i.e., the media bias problem. Bias of the news media is an inherent flaw of the news production process, spanning news gathering, writing, and editing stages. At every single stage, news is probably never free from the producer's subjective valuation and external forces such as owners and advertisers. While the problem has been extensively studied in the area of mass communication and journalism, effective solutions are barely developed. Our research investigates the media bias problem from a computational perspective and proposes a practical approach 'media bias mitigation'. Admitting the prevalence of bias, the approach attempts to reduce the effects of bias rather than to prove or correct it. Our work aims to provide readers with tools for active interaction with which they easily discover and compare diversity of existing biased views. We provide an overview of the works dealing with three important news article domains: NewsCube system supporting aspect-level browsing of straight news articles; Commenter's sentiment pattern-based analysis of political news articles; and Disputant relation-based method for news articles of contentious news issues. http://research.microsoft.com/apps/video/default.aspx?id=159104 Souneil Park Science and Technology Thu, 16 Feb 2012 22:00:00 GMT [Speaker: Bolin Ding] This talk is about index structures and data models to support efficient and effective keyword search, from a micro level (fundamental-operator level), e.g., data structures for fast set intersection, to a macro level (system level), e.g., leveraging taxonomies in keyword search. In the first part of the talk, I will discuss algorithms to improve the performance of a key operation in keyword search for both structured and unstructured data: the efficient computation of set intersections. Here, I will introduce compact data structures to represent sets such that their intersection can be computed in a worst-case efficient way. In general, given k (preprocessed) sets, with totally n elements, I will show how to compute their intersection in expected time O(n/sqrt(w) + kr), where r is the intersection size and w is the number of bits in a machine-word. Subsequently, I will discuss a much more practical version of this algorithm, which has weaker asymptotic guarantees but is much simpler and performs even better in practice; both algorithms outperform the state of the art techniques for both synthetic and real data sets and workloads. In the second part, I will describe how to optimize (inverted) indexes to support processing a type of keyword query that allows word substitutions defined by a given taxonomy of terms. This problem is challenging because each term in a query can have a large number of substitutions and the original query can be rewritten into any of their combinations. We propose to build an additional index (besides inverted index) to efficiently process these types of queries. For a given query workload, we formulate an optimization problem which chooses the additional index structure, aiming at minimizing the query evaluation cost, under given index space constraints. We show the NP-hardness of the problem, and propose a pseudo-polynomial time algorithm using dynamic programming, as well as an (1−1/e)/4-approximation algorithm to solve the problem. Experimental results show that, with only 10% additional index space, our approach can greatly reduce the query evaluation cost. Finally, I will briefly introduce our Text Cube-TopCell project which aims to support keyword queries and text analytics in structured data with both structural dimensions and text dimensions (or attributes). In the text cube model, text data is aggregated on different subsets of the structural dimensions, forming a hierarchy of entities called cube cells. Given a keyword query, our goal is to find the top-k most relevant cells in the text cube, which correspond to hidden entities at different granularities. Efficient algorithms to find relevant cube cells are proposed and a real system has been implemented (supported by NASA). http://research.microsoft.com/apps/video/default.aspx?id=159105 Bolin Ding Science and Technology Thu, 16 Feb 2012 21:30:00 GMT [Speaker: Sebastian Seung] Science has long struggled to pinpoint where, precisely, our uniqueness resides. A connectome is a map of connections between a brain’s neurons, and connectomics is the use of brain imaging and technology to increase the speed and efficiency of those maps. Your brain contains a million times more connections than your genome has letters. Finding the complete neuronal connectome of a human brain is one of the greatest scientific and technological challenges of all time. Seung’s goal is to compare connectomes between normal brains and disordered brains, which would reveal what’s behind brain disorders. If he and his team succeed, it could reveal the basis of personality, intelligence, memory, and mental disorders. Many scientists speculate that people with anorexia, autism, and schizophrenia are “wired differently.” This research has the potential to completely rock our understanding of the brain. http://research.microsoft.com/apps/video/default.aspx?id=159015 Sebastian Seung Science and Technology Thu, 16 Feb 2012 21:30:00 GMT [Speaker: Aditya Bhaskara] A basic primitive in graph optimization is that of finding small induced subgraphs of a given graph with “many” edges. Questions of this nature arise often in practice, for instance in finding “communities” in social networks, and in clustering type problems. From the theory side, it captures questions such as “small-set expansion”, which has proved to be an intriguing open problem. A simple formalization I will talk about is the Densest-k-subgraph problem, where the objective, given a graph G, is to find an induced subgraph on k vertices with as many edges as possible. There is a vast gap in our understanding of this problem in terms of the approximation ratio we can achieve. I will discuss the state of the art in approximation algorithms, and try to point out why it is difficult to write convex programming relaxations for the problem. This is akin to difficulties that arise in other problems which have a “small-support” constraint, such as those arising in compressed sensing. An interesting aspect of the densest-k-subgraph problem is that “random instances” seem to be the hardest for obtaining approximation algorithms. I will then talk about related “continuous” optimization questions, such as q-p norms of matrices, singular values of matrices and tensors, and discuss potential future directions. http://research.microsoft.com/apps/video/default.aspx?id=159016 Aditya Bhaskara Science and Technology Thu, 16 Feb 2012 19:00:00 GMT [Speaker: Filippo Lanubile] Global software teams usually rely on a Collaborative Development Environment (CDE) to cope with distance. In this seminar we present SocialTFS, an extension of Team Foundation Server (TFS), which aggregates information from social networks and presents it within Visual Studio. We argue that conveying social information into the developer’s workspace can work as a surrogate of the social awareness that we acquire during informal meetings and then help to build trust among members of global teams. Work on SocialTFS is in progress and visible on Codeplex. The project is funded by Microsoft Research Software Engineering Innovation Foundation (SEIF) Award for 2011. http://research.microsoft.com/apps/video/default.aspx?id=159017 Filippo Lanubile Science and Technology Thu, 16 Feb 2012 18:30:00 GMT [Speaker: Ravishankar Krishnaswamy] In this talk we will present approximation algorithms (and general techniques) for some basic problems in the field of stochastic optimization. A canonical problem is stochastic knapsack: we are given a knapsack of size B, and a collection of items, where the items have stochastic sizes and rewards (the distributions are known in advance). The goal is to devise an algorithm (i.e., a policy which can adapt based on the realized sizes of previously inserted items) to insert the next item to maximize the expected reward of items successfully packed into the knapsack.The size and reward of the item are realized only after the item is inserted. This basic problem can be generalized along several different directions. For example, if each of the stochastic jobs is more generally a Markov Chain (and makes random transitions each time we "play" it), we can then capture a class of (finite-horizon) multi-armed bandit problems. Or if these jobs are located at different vertices on a metric, we obtain stochastic routing and orienteering problems. While these problems have all received much attention in fields like operations research and machine learning, the issue of their approximability has been addressed only recently. In this talk, we will present efficient approximation algorithms with near-optimal guarantees for the (general) stochastic knapsack problem, the stochastic orienteering problem, and the multi-armed bandits problem. To the best of our knowledge, prior work (and techniques) only handle very special cases of the above problems. These results are based on joint works with Anupam Gupta, Marco Molinaro, Viswanath Nagarajan, and R. Ravi, and appeared at the FOCS 2011 and SODA 2012 conferences. Towards the end, I will also briefly discuss some of my other research in areas such as network design and scheduling. Finally, I will conclude by discussing the future directions in the above areas. http://research.microsoft.com/apps/video/default.aspx?id=159106 Ravishankar Krishnaswamy Science and Technology Wed, 15 Feb 2012 18:30:00 GMT [Speaker: Raymond W. Yeung] Network coding can significantly improve the transmission rate of communication networks with packet loss compared with routing. However, using network coding usually incurs higher computational and storage costs in the network devices and terminals. For example, some network coding schemes require the computational and/or storage capabilities of an intermediate network node to increase linearly with the number of packets for transmission, making them difficult to be implemented in a router-like device that has only constant computational and storage capabilities. In this talk, we introduce BATS code, which enables a digital fountain approach to resolve the above issue. BATS code is a coding scheme that consists of an outer code and an inner code. The outer code is a matrix generation of a fountain code. It works with the inner code which comprises random linear coding at the intermediate network nodes. BATS codes preserve such desirable properties of fountain codes as ratelessness and low encoding/decoding complexity. The computational and storage capabilities of the intermediate network nodes required for applying BATS codes are independent of the number of packets for transmission. It has been verified theoretically for certain special cases and demonstrated numerically for general cases that BATS codes can achieve rates very close to optimality. http://research.microsoft.com/apps/video/default.aspx?id=158951 Raymond W. Yeung Science and Technology Mon, 13 Feb 2012 21:30:00 GMT [Speaker: Jeff Hancock] Views of how social media can affect people are often over-generalized ("its ruining our conversational skills!" versus "its going to spark revolutions to rid the world of dictators!"), as is often the case with new technologies. The degree to which social media are embedded in our everyday lives makes this impulse all the more powerful. In the present talk, we'll go over some of the most recent research from our group that examines how social media can have subtle but powerful effects on a range of psychological dynamics. The talk will cover how emotional contagion can take place over text-based communication, how social media can boost self-esteem and self-affirmation while degrading performance on a math test, and modify the way we lie (and tell the truth) to one another. The talk will include several activities designed to illuminate the key principles derived from the experimental findings that I will present, so be prepared to have your emotions infected, your sense of self improved, and to be lied to. http://research.microsoft.com/apps/video/default.aspx?id=158952 Jeff Hancock Science and Technology Fri, 10 Feb 2012 21:30:00 GMT [Speaker: Hossein Mobahi] We investigate a well-known heuristic for optimization of nonconvex functions. The idea is to start from a highly smoothed version of the objective function that "hopefully" makes it convex (easy to globally minimize). This minimizer then initiates a gradient descent loop which follows the minimizer path while the objective function gradually deforms back to the original. This heuristic sometimes converges to the global minimum of the original function. The smoothing process is closely related to some fundamental processes in physics such as heat equation and schrodinger's equation. Yet, surprisingly, there is almost no theoretical understanding about this heuristic when used for global optimization. In the first part of the talk (perhaps of interest to a broad spectrum of audience), I present some results in these directions which include answers to the following questions: 1. What conditions guarantee a function is asymptotically (infinite smoothed) convex? 2. Is there any closed form for asymptotic minimizers? 3. How big is the class of asymptotically convex functions? 4. I also discuss some preliminary ideas and plans for developing performance guarantees for this optimization approach. Second part of the talk shows how these ideas can be applied to a fundamental problem in computer vision, namely the image alignment. We prove that the traditional lucas-kanade type alignment is only optimal when geometric transform is limited to displacement. We then discuss the optimal way to smooth the alignment objective for more general transformation models. We introduce a new concept called "transformation kernel" which allows efficient computation of the smoothed objective for common transformation models such as affine and homography. We show that smoothing the objective function outperforms lucas-kanade type of alignment. http://research.microsoft.com/apps/video/default.aspx?id=158953 Hossein Mobahi Science and Technology Fri, 10 Feb 2012 18:30:00 GMT [Speaker: Jason Short] Current Event: Managing data in your app How do you manage the data in your app? In this talk, Jason will present strategies for managing the data on the phone. He will cover the following: Loading initial data that will be replaced during runtime Strategies for storing data Data filtering and loading for UI binding Data expiration Speaker bio: Jason Short joined Microsoft in 2010 to work on the SQL Server Manageability team. Prior to joining Microsoft, Jason was an ISV for over 13 years. He has built systems as diverse as roller coasters, pyro controllers, medical imaging systems, and mobile entertainment. Please use the attached meeting invitations (open and accept them so we get attendee count). Note that “online only” is for those of you who can’t join us in building 99. Agenda: 6:00 PM – Arrive, mingle, get food 6:30 PM – Presentation 7:30 PM – Mingle, Q&A, etc. 8:00 PM – Event ends We will have presentations on the second Thursday of every month. http://research.microsoft.com/apps/video/default.aspx?id=158954 Jason Short Science and Technology Fri, 10 Feb 2012 02:00:00 GMT [Speaker: Adam Lashinsky] It has been said about Apple that its business practices are like a bumble bee: It shouldn’t fly, but it does. And how well it does. Apple is the first or second most valuable company in the world, and it got that way by doing business differently from how it is taught at Harvard Business School. The whole world loves Apple products, but even sophisticated business people don’t understand how Apple does what it does. Apple’s approach to leadership, personnel, secrecy, design, product development, marketing, public relations is extraordinarily unique approaches to business. http://research.microsoft.com/apps/video/default.aspx?id=158955 Adam Lashinsky Science and Technology Thu, 09 Feb 2012 23:00:00 GMT [Speaker: Qi Tian] Bag-of-visual-words model is widely used in the state-of-the-art large-scale image retrieval system. It represents each image as a bag of visual words by quantizing local image descriptors to the closest visual words. However, feature quantization reduces the discriminative power of local features, which causes many false visual word matches. Recently, some geometric verification methods are proposed to check the geometric consistency of matched features in a post-processing step. Although retrieval precision is improved, either the computational cost is too expensive to ensure real-time response, or they are limited to local verification. To address this dilemma, we propose a novel scheme, Spatial Coding, designed for large scale partial-duplicate image retrieval. The spatial relationships among visual words are encoded in global region maps. Based on the region maps, a spatial verification approach is developed, which can detect false matches of local features efficiently, and consequently improve retrieval performance greatly. Experiments in partial-duplicate image retrieval, using a database of one million images from Image-Net, reveal that our approach can effectively detect duplicate images with rotation, scale changes, occlusion, and background clutter with very low computational cost. The spatial coding achieve an 53% improvement in mean average precision and 46% reduction in time cost over the baseline Bag-of-Visual-Words approach, respectively. They perform even better than full geometric verification while being much less computationally expensive. Our demo on 10-million dataset further reveals the scalability of our approach. http://research.microsoft.com/apps/video/default.aspx?id=158956 Qi Tian Science and Technology Thu, 09 Feb 2012 18:30:00 GMT [Speaker: Vagelis Hristidis] The popularity of collaborative tagging sites has created new challenges and opportunities for designers of web items, such as electronics products, travel itineraries, popular blogs, etc. An increasing number of people are turning to online reviews and user-specified tags to choose from among competing items. This creates an opportunity for designers to build items that are likely to attract desirable tags when published. We consider a novel optimization problem: given a training dataset of existing items with their user-submitted tags, and a query set of desirable tags, design the k best new items expected to attract the maximum number of desirable tags. We present algorithms for solving this problem and present experimental results. http://research.microsoft.com/apps/video/default.aspx?id=158957 Vagelis Hristidis Science and Technology Wed, 08 Feb 2012 18:30:00 GMT [Speaker: Rebecca MacKinnon] In an age where technology companies censor and turn over users’ personal information at the behest of government, internet policy expert Rebecca MacKinnon argues a different approach to protecting our personal liberties and freedoms. In Consent of the Networked, MacKinnon proposes focusing policy upstream—at the point of conception and innovation—to ensure accountability is built into the fabric of cyberspace. http://research.microsoft.com/apps/video/default.aspx?id=158668 Rebecca MacKinnon Science and Technology Tue, 07 Feb 2012 21:30:00 GMT [Speaker: Sukhada Fadnavis] The birthday problem states that there is at least half a chance that some two out of twenty-three randomly chosen people will share the same birth date. The calculation for this problem assumes that all birth dates are equally likely. What if the distribution of birth dates is non-uniform and possibly even unknown? Further what if we focus on birthdays shared by two friends rather than any two people? I will present some of our results and conjectures in this generalized setting. I will also show how these results are related to the Stanley-Stembridge poset chain conjecture and the 'shameful conjecture', two famous conjectures in combinatorics. http://research.microsoft.com/apps/video/default.aspx?id=158669 Sukhada Fadnavis Science and Technology Fri, 03 Feb 2012 18:30:00 GMT [Speakers: Rob Fatland and Shoshanna Budzianowski] 8:30 to 10:30 New Collaborations At this meeting Microsoft Research will present details of projects they are undertaking which have relevance to the Integrated Ocean Observing System. This will include a summary of their recent work on the development of ‘Eye on Earth’ (http://watch.eyeonearth.org); a portal for public delivery of environmental data and the incorporation of data from citizen observing networks. The meeting will provide an opportunity to discuss and develop new ideas for distribution and use of marine data and the potential for public contribution of observations. 11:00 to 1:00 Enhancing Observations A meeting aimed at further developing the outcomes of the previous day’s workshop and determining how to take them forward. In particular, this meeting will seek to address three main themes: Achieving optimum integration of the Pacific Northwest regional observing efforts and needs of the Federal Agencies represented on the Interagency Ocean Observation Committee; Encouraging wider participation of stakeholders in the work of NANOOS; Securing and sustaining improved engagement with all end-users of marine observations and forecasts. http://research.microsoft.com/apps/video/default.aspx?id=158592 Rob Fatland; Shoshanna Budzianowski Science and Technology Fri, 03 Feb 2012 16:30:00 GMT [Speaker: James Waller] The past century, dubbed the "Age of Genocide,” saw more than 60 million people murdered to meet the needs of the state. One unassailable fact behind this litany of human conflict and suffering is that political, social, or religious groups wanting to commit mass murder are never hindered by a lack of willing executioners. How is it that ordinary people like you and me, commit such extraordinary evil? http://research.microsoft.com/apps/video/default.aspx?id=158670 James Waller Science and Technology Fri, 03 Feb 2012 02:30:00 GMT [Speakers: Hans R. Moritz, Amy MacFadyen, David Michalsen, Dan Jordan, and Micheal Schoonover] 4:10 to 4:25Case Study Example – Coastal Engineering Hans R. Moritz, United States Army Corps of Engineers (potentially also Joan Oltman-Shay, Northwest Research Associates) This case study will focus on coastal infrastructure design and describe how observations have been used in risk assessment and analysis 4:25 to 4:35Discussion – Linking needs to capabilities 4:35 to 4:50Case Study Example – Hazard Response and Marine Operations (invited) United States Coast Guard Amy MacFadyen, NOAA HAZMAT This case study will provide an understanding of real-time data needs from the perspective of emergency response and marine operations. 4:50 to 5:00Discussion – Linking needs to capabilities SUMMING UP AND NEXT STEPS Jan Newton, Executive Director of NANOOS 5:00 to 5:30A synthesis of workshop outcomes. http://research.microsoft.com/apps/video/default.aspx?id=158958 Hans R. Moritz; Amy MacFadyen; David Michalsen; Dan Jordan; Micheal Schoonover Science and Technology Fri, 03 Feb 2012 00:10:00 GMT [Speakers: Peter Lawson, Joe Schumacker, Alan Pazar, Michael Kosro, Allan Barton, Andy Suhbier, Benoir Eudeline, Belinda Batten, and Brian Polagye] 2:05 to 2:45Case Study Example – Fisheries Peter Lawson, National Oceanic & Atmospheric Administration Joe Schumacker, Quinault Indian Nation This case study will detail the local fisheries use of observations and derived products for improved decision-making and opportunities for enhanced observing. 2:45 to 2:55Discussion – Linking needs to capabilities 2:55 to 3:10Case Study Example – Aquaculture (invited), Allan Barton, Whiskey Creek Hatchery Andy Suhbier, Pacific Shellfish Institute This case study will provide an understanding of how measurements play a role in operational decision-making for the aquaculture industry. 3:10 to 3:20Discussion – Linking needs to capabilities 3:20 to 3:35Case Study Example – Alternative Energy Belinda Batten, OSU, Northwest National Marine Renewable Energy Center Brian Polagye, UW, Northwest National Marine Renewable Energy Center This presentation will look at how observations are used in support of selecting wave and tidal locations and operations. 3:35 to 3:45Discussion – Linking needs to capabilities http://research.microsoft.com/apps/video/default.aspx?id=158959 Peter Lawson; Joe Schumacker; Alan Pazar; Michael Kosro; Allan Barton; Andy Suhbier; Benoir Eudeline; Belinda Batten; Brian Polagye Science and Technology Thu, 02 Feb 2012 22:05:00 GMT [Speaker: Susan Cain] At least one-third of the people we know are introverts. They are the ones who prefer listening to speaking, reading to partying; who innovate and create but dislike self-promotion; who favor working independently over brainstorming in teams. Although they are often labeled "quiet," it is to introverts that we owe many of the great contributions to society--from van Gogh’s sunflowers to the invention of the personal computer. The dominant values of American business culture, where forced collaboration can stand in the way of innovation, and where the leadership potential of introverts is often overlooked. Quiet shows how dramatically we undervalue introverts, and how much we lose in doing so. http://research.microsoft.com/apps/video/default.aspx?id=158430 Susan Cain Science and Technology Thu, 02 Feb 2012 21:30:00 GMT