Tuesday, October 30, 2007
9:25 - 10:30
Swarming the Future
Data Management Systems for Monitoring Applications
Extracting Information from the Web
10:40 - 11:30
Computing for Everyone - Accessible Computing
Computer Architecture is Cool Again
The Science Behind Advertising on the Web
11:40 - 12:30
Security, Privacy, and the Internet
HCI Meets Personal Activity Modeling
- 9:25-9:30: Introduction and Overview: Tom Anderson
- 9:30-9:45: A Case for OneSwarm, Michael J. Piatek (PDF Slides)
Peer-to-peer data sharing systems offer the potential for vastly improved scalability, robustness, and efficiency for modern many-to-many communication patterns dominating the Internet today. To realize this potential, we need to design a robust incentive strategy to encourage users to contribute resources instead of free-riding and to encourage ISPs to assist P2P systems rather than fight them. We conduct and report a set of measurements of live BitTorrent swarms, finding that persistent incentives spanning interactions across objects are needed to address these issues. We use these insights to sketch the design of a system called OneSwarm and outline some remaining open research questions.
- 9:45-10:00: Scalable DoS Protection with Phalanx, Colin Dixon (PDF Slides)
Large-scale distributed denial of service (DoS) attacks are an unfortunate everyday reality on the Internet. They are simple to execute and with the growing prevalence and size of botnets more effective than ever. Although much progress has been made in developing techniques to address DoS attacks, no existing solution is unilaterally deployable, works with the Internet model of open access and dynamic routes, and copes with the large numbers of attackers typical of today's botnets.
In this paper, we present a novel DoS prevention scheme to address these issues. Our goal is to define a system that could be deployed in the next few years to address the danger from present-day massive botnets. The system, called Phalanx, leverages the power of swarms to combat DoS. Phalanx makes only the modest assumption that the aggregate capacity of the swarm exceeds that of the botnet. A client communicating with a destination bounces its packets through a random sequence of end-host mailboxes; because an attacker doesn't know the sequence, they can disrupt at most only a fraction of the traffic, even for end-hosts with low bandwidth access links. We use PlanetLab to show that this approach can be both efficient and capable of withstanding attack.
- 10:00-10:15: Finding Internet Black Holes with Hubble, Ethan Katz-Bassett (PDF Slides)
We present Hubble, a system that operates continuously to find Internet blackholes in which routes exist to a destination but packets are unable to reach the destination. Hubble allows us to characterize global Internet reachability by identifying how many prefixes are reachable from some vantages and not others, how often these problems occur, and how long they persist. It also classifies the problems in ways that aid troubleshooting. Hubble monitors the Internet's edge address space at a 15 minute granularity. Key enabling techniques include a hybrid passive/active monitoring approach and the synthesis of multiple information sources that include historical data. With these techniques, we estimate that Hubble discovers 85% of the reachability problems that would be found with a pervasive probing approach, while issuing only 5.5% as many probes.
We also present the results of a three week study conducted with Hubble. We find that the extent of reachability problems, both in number and duration, is much greater than we expected, with problems persisting for hours and even days, and many of the problems do not correlate with BGP updates. In many cases, a multi-homed AS is reachable through one provider, but probes through another terminate; using spoofed packets, we isolated the direction of failure in three-fourths of cases we analyzed and found all these problems to be exclusively caused by the provider not forwarding traffic to the destination.
- 10:15-10:30: Wireless Interference Cancellation, Dan Halperin (PDF Slides)
We argue that carrier sense in 802.11 and other wireless protocols makes scheduling decisions that are overly pessimistic and hence waste capacity. As an alternative, we suggest interference cancellation, in which simultaneous signals are modeled and decoded together rather than treating all but one as random noise. This method greatly expands the conditions under which overlapping transmissions can be successfully received, even by a single receiver. We demonstrate the practicality of these better receivers via a proof-of-concept experiment with USRP software radios.
This type of receiver enables new and more effective wireless MACs in which more aggressive scheduling decisions can be made, increasing wireless capacity in a multiuser environment. Additional benefits of these techniques include increased robustness of wireless networks to external devices.
- 9:25-9:30: Introduction and Overview: Magdalena Balazinska
- 9:30-9:45: Extracting Events from Imprecise Data Streams, Christopher Re
In recent years, database researchers have developed event detection systems, which provide a query language and runtime that allows users to monitor and detect complex events in streams of data. However, current systems assume that data are precise, which fails in many real world situations. For example, missed RFID readings create imprecision about the location of a tag. In this talk, we describe our system PEEX+ (Probabilistic Event Extraction+) which allows users to extract events on imprecise streams. In general, processing uncertain data is intractable (NP-hard), however we find that some queries -- including all regular expressions -- can be processed efficiently on probabilistic streams.
- 9:45-10:00: Complex Event Specification for Non-Expert Users of RFID Data, Evan Welbourne
A key challenge in pervasive computing is the transformation of sensor data into higher-level events (e.g., Bob is getting coffee, Alice is printing a paper). Emerging event detection systems can provide suitably flexible services for extracting these high-level events, but they require event specifications in complex, expert-oriented languages. We present the design and implementation of SCENIC, a tool that allows end users to graphically specify spatio-temporal event definitions using an intuitive iconic language and storyboard metaphor. SCENIC automatically translates user-specified events into an event language used by a sophisticated event detection engine. We show that non-experts can use SCENIC to generate event specifications which produce meaningful results when run over real experimental data collected in an RFID-based pervasive computing environment.
- 10:00-10:15: Privacy Issues and Techniques for Monitoring Applications, Vibhor Rastogi (PDF Slides)
Many real world monitoring applications collect personal information about individuals, such as their location or their activities, to provide new types of services including reminders, people or object finders, and smart diaries. This information collection raises many privacy concerns. In this talk, we highlight some of the privacy issues that arise in the RFID Ecosystem, an RFID-based computing system that collects location information from mobile users and their belongings. We illustrate how the uncertainty in the collected location information further compounds the privacy problem. As a first step, we propose an approach that combines database privacy methods like perturbation with techniques used in probabilistic databases to handle the uncertainty in the location information.
- 10:15-10:30: Moirae: History-Enhanced Monitoring, YongChul Kwon (PDF Slides)
Today, many applications process continuous streams of information in near real-time. These applications include financial services, network monitoring, computer system monitoring, and sensor-based environment monitoring. These applications also archive the data streams they process for reasons including audit, billing, quality assurance, and forensic analysis. This archived data is an abundant source of information that may help improve the quality of the real-time monitoring tasks. Existing stream processing systems, however, provide limited support for exploiting stream data archives. The goal of the Moirae project is to provide a new type of stream processing engine that integrates near real-time information with data archives to enable new types of smart monitoring tasks. In this talk, we present the current status and future goals of the Moirae project.
- 9:25-9:30: Introduction and Overview: Dan Weld
- 9:30-9:45: Assieme: Finding and Leveraging Implicit References in a Web Search Interface for Programmers, Raphael Hoffmann (PDF Slides)
Programmers regularly use search as part of the development process, attempting to identify an appropriate API for a problem, seeking more information about an API, and seeking samples that show how to use an API. However, neither general-purpose search engines nor existing code search engines currently fit their needs, in large part because the information programmers need is distributed across many pages. We present Assieme, a Web search interface that effectively supports common programming search tasks by combining information from Web-accessible Java Archive (JAR) files, API documentation, and pages that include explanatory text and sample code. Assieme uses a novel approach to finding and resolving implicit references to Java packages, types, and members within sample code on the Web. In a study of programmers performing searches related to common programming tasks, we show that programmers obtain better solutions, using fewer queries, in the same amount of time spent using a general Web search interface.
- 9:45-10:00: Information Extraction for Structured Web Queries, Michael J. Cafarella (PDF Slides)
The Web contains a huge amount of text that is currently beyond the reach of structured access tools. This unstructured data often contains a substantial amount of implicit structure, much of which can be captured using information extraction (IE) algorithms. By combining an IE system with an appropriate data model and query language, we could enable structured access to all of the Web's unstructured data. I will discuss possible applications and technical challenges in accessing this data, motivated by experiences with a 90M-page prototype.
- 10:00-10:15: From Wikipedia to the Semantic Web, Fei Wu (PDF Slides)
Berners-Lee's compelling vision of a Semantic Web is hindered by a chicken-and-egg problem, which can be best solved by a bootstrapping method -- creating enough structured data to motivate the development of applications. This paper argues that autonomously "Semantifying Wikipedia" is the best way to solve the problem. We choose Wikipedia as an initial data source, because it is comprehensive, not too large, high-quality, and contains enough manually-derived structure to bootstrap an autonomous, self-supervised process. We identify several types of structures which can be automatically enhanced in Wikipedia (e.g., link structure, taxonomic data, infoboxes, etc.), and we describe a prototype implementation of a self-supervised, machine learning system which realizes our vision. Preliminary experiments demonstrate the high precision of our system's extracted data -- in one case equaling that of humans.
- 10:15-10:30: Joint Inference in Information Extraction, Hoifung Poon (PDF Slides)
The goal of information extraction is to extract database records from text or semi-structured sources. Traditionally, information extraction proceeds by first segmenting each candidate record separately, and then merging records that refer to the same entities. While computationally efficient, this approach is suboptimal, because it ignores the fact that segmenting one candidate record can help to segment similar ones. For example, resolving a well-segmented field with a less-clear one can disambiguate the latter's boundaries. In this paper we propose a joint approach to information extraction, where segmentation of all records and entity resolution are performed together in a single integrated inference process. While a number of previous authors have taken steps in this direction ( e.g., Pasula et al (2003), Wellner et al. (2004)), to our knowledge this is the first fully joint approach. In experiments on the CiteSeer and Cora citation matching datasets, joint inference improved accuracy, and our approach outperformed previous ones. Further, by using Markov logic and the existing algorithms for it, our solution consisted mainly of writing the appropriate logical formulas, and required much less engineering than previous ones.
Swarming the Future (CSE 305)
Data Management Systems for Monitoring Applications (CSE 403)
Extracting Information from the Web (CSE 691)
- 10:40-10:45: Introduction and Overview: Richard Ladner
- 10:45-11:00: Improving the Performance of Motor-Impaired Users with Automatically-Generated, Ability-Based Interfaces, Krzysztof Gajos (PDF Slides)
Most of today's GUIs are designed for the typical, able-bodied user; atypical users are, for the most part, left to adapt as best they can, perhaps using specialized assistive technologies as an aid. In this paper, we present an alternative approach: SUPPLE++ automatically generates interfaces which are tailored to an individual's motor capabilities and can be easily adjusted to accommodate varying vision capabilities. SUPPLE++ models users' motor capabilities based on a onetime motor performance test and uses this model in an optimization process, generating a personalized interface.
In a study comparing this approach to baseline interfaces, our results show that users with motor impairments were much faster and strongly preferred SUPPLE++ ability-based interfaces. Specifically, motor-impared participants were 26.4% faster using interfaces generated by SUPPLE++. They made 73% fewer errors, strongly preferred those interfaces to the manufacturers’ defaults, and found them more efficient, easier to use, and much less physically tiring. These findings indicate that rather than requiring some users with motor impairments to adapt themselves to software using separate assistive technologies, software can now adapt itself to the capabilities of its users.
Joint work with Jacob Wobbrock, Daniel Weld
- 11:00-11:15: A Web 2.0 Approach to Improving Web Accessibility, Jeffrey Bigham (PDF Slides)
Browsing the web is inefficient for blind web users because of persistent accessibility problems. Web 2.0 presents new challenges for accessibility, but also offers new opportunities both for understanding how blind people use the web and for enabling new tools to improve access to web content. In this talk, we first present the results of a study that we conducted indicating dramatic differences in the normal browsing behavior of blind and sighted web users. This study quantitatively showed that blind users avoid visiting highly-dynamic web pages and avoid interacting with content that has not been optimized for accessibility. Next, we present the design of a screen-reading web application that provides full web access to blind web users from any computer without requiring a traditional $1000 screen reader to be installed. This application could provide access to the web for blind people who are either on-the-go or who are unable to afford a pricey screen reader. It could also serve as a convenient, low-cost tool for web developers to check the accessibility of the web pages they produce.
- 11:15-11:30: The Deaf and Hard of Hearing Cyber-Community in Science, Technology, Engineering, and Mathematics, Richard Ladner
The national cyberinfrastructure has the potential to better facilitate the inclusion and advancement of deaf and hard of hearing students in the advanced Science, Technology, Engineering, and Mathematics (STEM) fields. Succeeding in mainstream universities (at all levels) involves extra challenges for deaf and hard of hearing students: skilled sign language interpreters and captioners with advanced domain knowledge are most often difficult to find; multiple visual channels of information in the classroom can be difficult to juggle; and collaboration inside and outside the classroom is often strained due to language barriers. Furthermore, translation of advanced STEM topics into American Sign Language (ASL) is far from standardized and often requires discussion between students and interpreters to devise consistent signs. Better access to classroom activities and a consistent, conceptually clear signing system for STEM topics are both vitally needed in order for deaf and hard of hearing students to advance in the sciences. We describe the new research program to (i) enable access to STEM education by designing, building and evaluating flexible educational technologies that support remote sign language interpreting and captioning and (ii) enable ASL to grow in STEM by designing, building, and evaluating a sign language social networking forum that supports video up-load and discussion of signs in STEM.
- 10:40-10:45: Introduction and Overview: Mark Oskin
- 10:45-11:00: How to Make Multiprocessors Less Scary, Luis Ceze
Multicores are the architecture of choice (arguably the only truly viable one) for scaling on-chip computing resources. However, both the hardware and the software industries are scared of multicores. The hardware industry relies on the software industry to motivate customers to buy their newest multicore chips; without that software driver, chip sales will decline. However, to provide consumers with the need for multicore chips, the software industry faces the daunting task of writing code that benefits from medium-scale parallel computing.
In this talk, I will explain my view on why multiprocessors are so scary. In doing that, I will talk about our current research efforts in making multiprocessor architectures less of a nightmare to program. These efforts include making the system: (1) more resilient to concurrency bugs, (2) easier to debug, by decreasing non-determinism, and (3) easier to understand, by supporting simpler consistency models without sacrificing performance.
- 11:00-11:15: Architectural Implications of Brick and Mortar Silicon Manufacturing, Martha Kim (PDF Slides)
We introduce a chip fabrication technique called "brick and mortar", in which chips are made from small, pre-fabricated ASIC bricks and bonded in a designer-specified arrangement to an inter- brick communication backbone chip. The goal is to provide a low- overhead method to produce custom chips, yet with performance that tracks an ASIC more closely than an FPGA. We have examined the architectural design choices in this chip-design system. These choices include the definition of reasonable bricks, both in functionality and size, as well as the communication interconnect that the I/O cap provides. We discuss a sample chip design, a 16-way CMP, and analyze the costs and benefits of designing chips with brick and mortar. We find that this method of producing chips incurs only a small performance loss (8%) compared to a fully custom ASIC, which is significantly less than the degradation seen from other low- overhead chip options, such as FPGAs. Finally, we measure the effect that architectural design decisions have on the behavior of the proposed physical brick assembly technique, fluidic self-assembly.
- 11:15-11:30: Designing a Coarse-Grained Reconfigurable Architecture for Power Efficiency, Brian C. Van Essen (PDF Slides)
Coarse-grained reconfigurable architectures (CGRAs) have the potential to offer performance approaching an ASIC with the flexibility, within an application domain, similar to a digital signal processor. In the past, coarse-grained reconfigurable architectures have been encumbered by challenging programming models that are either too far removed from the hardware to offer reasonable performance or bury the programmer in the minutiae of hardware specification. Additionally, the ratio of performance to power hasn't been compelling enough to overcome the hurdles of the programming model to drive adoption.
The goal of our research is to improve the power efficiency of a CGRA at an architectural level, with respect to a traditional island-style FPGA. Additionally, we are continuing previous research into a unified mapping tool that simplifies the scheduling, placement, and routing of an application onto a CGRA.
- 10:40-11:00: Some Game Theoretic and Algorithmic Issues in Online Advertising, Anna Karlin (PDF Slides)
The online advertising industry is growing by leaps and bounds. For example, in the first half of 2007, Internet ad revenue totaled almost $10 billion, an increase of nearly 27% from the same period a year ago.
In this talk, we give a high level survey of some of the algorithmic and game-theoretic issues relevant to the design of mechanisms for allocating and pricing online ads.
- 11:00-11:15: On Best-Response Bidding in Ad Auctions, Yiannis Giotis
A fundamental problem faced by advertisers bidding for slots in a keyword auction is how to bid so as to maximize their profit. In this talk, we describe and analyze a class of best-response bidding strategies for a repeated auction on a single keyword. These bidding strategies are simple and easily implementable, and also, as we show, achieve several desirable criteria for the advertisers. In the talk, we describe our theoretical and empirical results on the revenue, convergence and robustness properties of these strategies.
Joint work with M. Cary, A. Das, B. Edelman, K. Heimerl, A. Karlin, C. Mathieu and M. Schwarz
- 11:15-11:30: Viral Marketing: Algorithms and Impossibilites for Adoption of a New Product, Ning Chen
Viral marketing is the process by which a company tries to take advantage of word-of-mouth effects to market a product. The hope is that early adopters of a product will influence their friends to adopt the product, who in turn will influence their friends, and so on, creating a large wave of adoptions. The question then becomes, given an underlying social network describing relations of individuals, to whom should the company initially market the product, e.g., to whom should the company offer free samples? We consider a number of different variants of this optimization problem, and describe corresponding algorithmic and impossibility results.
Computing for Everyone - Accessible Computing (CSE 305)
Computer Architecture is Cool Again (CSE 403)
The Science Behind Advertising on the Web (CSE 691)
- 11:40-11:45: Introduction and Overview: Steve Gribble
- 11:45-12:00: The Organization and Sharing of Web-Service Objects with Menagerie, Roxana Geambasu (PDF Slides)
The radical shift from the PC desktop to Web-based services is scattering personal data across a myriad of Web sites, such as Google, Flickr, YouTube, MySpace, and Amazon S3. This dispersal poses significant new challenges for users, making it more difficult for them to: (1) organize, search, and archive their data, much of which is now hosted by Web sites; (2) create heterogeneous (multi-Web-service) object collections and share them in a protected way; and (3) manipulate Web objects with standard applications or build new tools or scripts that operate on those objects.
This talk presents Menagerie, a software framework that addresses these challenges. Menagerie creates an integrated file and object system from heterogeneous, personal Web-service objects dispersed across the Internet. Our Menagerie architecture has two key parts. The Menagerie Service Interface (MSI) defines a common Web-service API for object naming, protection, and access. The Menagerie File System (MFS) lets desktop applications and Web services manipulate remote Web objects as if they were local files. Our experience shows that Menagerie greatly simplifies the construction of new applications that support collections of heterogeneousWeb objects and fine-grained protected sharing of those objects.
- 12:00-12:15: Detecting In-Flight Page Changes with Web Tripwires, Charles Reis (PDF Slides)
- 12:15-12:30: Physical Access Control in the RFID Ecosystem, Travis Kriplean (PDF Slides)
RFID security is a vibrant research area, and many protection mechanisms against unauthorized RFID cloning and reading attacks are emerging. However, little work has yet addressed the complementary issue of privacy for RFID data after it has been captured and stored by an authorized system. In this talk, I discuss the problem of peer-to-peer privacy for personal RFID data. In this setting, we assume a system with trusted owners and administrators, and focus on ways to constrain peers' access to information about one another. An access control policy, called Physical Access Control, protects privacy by constraining the data a user can obtain from the system to those events that occurred when and where that user was physically present. PAC provides a high level of privacy, offering a database view that augments users' memory of places, objects, and people. PAC is a natural, intuitive access-control policy for peer-to-peer privacy. It enables many classes of applications while providing a good baseline trade-off between privacy and utility.
- 11:40-11:45: Introduction and Overview: James Landay
- 11:45-12:00: Activity-Based Applications, James Landay
Recent advances in small inexpensive sensors, low-power processing, machine learning, and mobile user interfaces have enabled applications that use on-body sensing to infer people's activities throughout everyday life. The Digital Simplicity project brings this technology into people’s everyday lives and makes the high-level, long-lived activities individuals are currently doing simpler. In this talk I will discuss our first two capstone applications, the UbiFit Garden, which uses this technology to encourage individuals to be physically active by making them aware of their current physical activities and encouraging them to do more, and UbiGreen, which makes people more aware of their transportation patterns and gives them positive feedback when these patterns have less impact on the environment.
- 12:00-12:15: Activity-Based Prototyping of Ubicomp Applications, Yang Li
Ubiquitous computing (ubicomp) promises to support our everyday activities by bringing computing power into the fabric of everyday life. However, it is challenging to design for the complex, dynamic real world in comparison to the well-understood, desktop environment. We designed an activity-based prototyping process realized in the ActivityDesigner system that combines the theoretical framework of Activity-Centered Design with traditional iterative design. This process allows designers to leverage human activities as first class objects for design and is supported in ActivityDesigner by three novel features. First, this tool allows designers to model activities based on concrete scenarios collected from people's everyday lives. The extracted models form a context for design and computational constructs for creating functional prototypes. Second, it allows designers to prototype interaction behaviors based on activity streams spanning time. Third, it allows designers to easily test these prototypes with real users continuously in a longitudinal, in situ manner. We have garnered positive feedback from a series of laboratory user studies and several case studies in which ActivityDesigner was used in realistic design situations. Based on our experience as well as that of others, ActivityDesigner was able to effectively streamline an ubicomp design process, and allowed creating realistic ubicomp application prototypes at a low cost and test them in people's everyday lives over an extended time period.
- 12:15-12:30: Sensing from the Basement: A Feasibility Study of Unobtrusive and Low-Cost Home Activity Recognition, James Fogarty (PDF Slides)
The home deployment of sensor-based systems offers many opportunities, particularly in the area of using sensor-based systems to support aging in place by monitoring an elder’s activities of daily living. But existing approaches to home activity recognition are typically expensive, difficult to install, or intrude into the living space. This paper considers the feasibility of a new approach that "reaches into the home" via the existing infrastructure. Specifically, we deploy a small number of low-cost sensors at critical locations in a home’s water distribution infrastructure. Based on water usage patterns, we can then infer activities in the home. To examine the feasibility of this approach, we deployed real sensors into a real home for six weeks. Among other findings, we show that a model built on microphone-based sensors that are placed away from systematic noise sources can identify 100% of clothes washer usage, 95% of dishwasher usage, 94% of showers, 88% of toilet flushes, 73% of bathroom sink activity lasting ten seconds or longer, and 81% of kitchen sink activity lasting ten seconds or longer. While there are clear limits to what activities can be detected when analyzing water usage, our new approach represents a sweet spot in the tradeoff between what information is collected at what cost.
- 11:40-11:45: Introduction and Overview: Steve Seitz
- 11:45-12:00: Scene Summarization for Online Photo Collections, Ian Simon, (PDF Slides)
How can the essence of Rome be captured in a few images? A search for the tag "rome" on online photo sharing site Flickr yields over 700,000 images, listed in no discernible order. The distribution of these images, however, contains useful information about the importance of Rome's various sights. An ideal summary presents the most important sights with minimal redundancy. We use the distribution of images from Flickr to select a set of canonical views to form the scene summary, using clustering techniques on visual features. The summaries we compute also lend themselves naturally to the browsing of image collections, and can be augmented by analyzing user-specified image tag data. In addition, our summaries can be incorporated into a 3D browser such as Photo Tourism to enable a user to quickly find interesting viewpoints.
- 12:00-12:15: Internet Stereo: Accurate 3D Shape from Photos on the Internet, Noah Snavely (PDF Slides)
The Internet has become an incredibly rich source of imagery, with billions of photos available through sites like Flickr. For nearly all well-known cities, buildings, tourist sites, and national parks, thousands or millions of photos exist and are readily accessible. In our previous work on Photo Tourism, we were able to take a set of Internet photos of such sites and determine where each photo was taken. In this project, we build on this work to use the Internet to replace a 3D laser scanner: our goal is to leverage this massive amount of data to reconstruct accurate, dense 3D models of these sites. The system we have developed is fully automatic: to reconstruct a scene, we simply gather Internet photos using a Flickr search query, and the system builds the 3D model. The main contribution of our work is a new multi-view stereo algorithm that is robust to the large variability -- in weather, time of day, and camera quality -- present in Internet photo collections. We have demonstrated our system on several scenes, including the Notre Dame cathedral, the Statue of Liberty, and the Venus de Milo.
- 12:15-12:30: Bridging the Gap Between Photographs and Videos, Pravin Bhat (PDF Slides)
The motivation for this project comes from the observation that we have a lot more agency over our photographs than our videos. This is precisely why the difference in quality between amateur and professional content on photo sharing sites like Flickr is a lot smaller than the difference observed on video sharing sites like YouTube. There are many practical reasons for this disparity between the two visual media. Video produces much more data than still photography and thus necessitates more aggressive compression and limited resolution. Still cameras can flash, while video has to make do with the available light in the scene; this lack of light is compounded by the short shutter time that video requires to avoid motion blur when the camera is moving or zooming. Finally, the presence of a temporal axis makes editing videos significantly harder than editing photographs. For example, changes made to the appearance of an object in a video have to be consistent over time, compensate for occlusions, and handle interactions with scene lighting which can give rise to complex phenomenon such as reflections, spectacular highlights, caustics, and shadows.
Our ongoing work on video enhancement brings the benefits of digital photography to video. Using our system users can automatically enhance their low-quality videos by associating them with a few high-quality photographs. This work also allows users to edit entire videos by simply editing a couple of video frames.