Intelligent scissors is an interactive image segmentation tool that allows a user to select piece-wise globally optimal contour segments (based on an optimal path search in a graph) that correspond to a desired object boundary. Our current work uses tobogganing to raise the granularity of the image primitive above the pixel level, producing a region-based basic processing unit that is object-centered rather than device-dependent. The resulting region-based elements form the basis several improvements to intelligent scissors that reduce (often greatly) the human time and effort needed for object selection while simultaneously increasing the accuracy of boundary definition. Included in these improvements is an edge confidence measure that allows for improved cursor snapping, automatic path extension, and the ability to select certain objects simply by snapping to them. Also, the tobogganed-based regions provide for an edge model that computes subpixel boundary position, noise-independent edge blur adjustment, and automatic alpha matte generation and color seperation of boundary transition pixels--thereby improving the quality of certain image editing operations such as cut and paste.

Biography

Dr. Eric Mortensen joined the Computer Science Department at Oregon State University in Aug. 2001 after completing a Ph.D. in Computer Science at Brigham Young University. During his graduate work, Dr.Mortensen developed and extended a "cutting edge" user-guided image segmentation techniques called intelligent scissors. In addition to image and video segmenation and editing, Dr. Mortensen's other research interests include image-based modeling and interactive vision and graphics techniques.

Personal Digital Assistants (PDAs), cell phones, wireless networks and instant messaging have opened new opportunities for collaborative ebusiness and enterprise computing. Software agents, personalization techniques, user interface technologies and integration architectures are effective building blocks for engineering scalable middleware solutions or "web-ware". It is now possible to envisage bolt-on web-ware solutions that incrementally extend existing applications for collaborating end-users and enterprise workers. This seminar provides an overview of an on-going web-ware project at The Wise Net Inc. and directly related R&D at OSU. The development and R&D teams are exploring the application of Jini/Javaspaces, SOAP, XML, constraint processing and security technologies to build collaborative middleware for the real estate industry. Technical challenges from the perspective of the software engineer will be presented. Current OSU R&D projects and problem areas needing to be addressed are outlined.

Biography

Kal Toth is an Associate Professor at Oregon State University in the Department of Computer Science. He has over 25 years of industry experience in the fields of software engineering, electronic commerce, project management, distributed information networks and information security. He is actively conducting research in the field of ebusiness investigating wireless and wire-line personalization, security and interoperability problems using intelligent software agents and web technologies.

Visualization research is the science and art of designing, implementing and evaluating methods for effectively communicating information through images. Fundamental questions include: How can we determine how best to portray a large, complicated set of data so that its essential features can be accurately and intuitively understood? How can we best measure the success of our efforts? Where should we look to gain insight into the science behind the art of effective visual representation? In this talk I will address these and other visualization research issues in the context of some of my recent work in designing effective textures for 3D shape representation and 2D multivariate flow visualization. In particular, I will describe our efforts to unite inspiration from art with insights from fundamental findings in human visual perception to define the characteristics of texture patterns that can be effectively used for visualization purposes such as facilitating accurate shape perception and the integrated comprehension of multivariate distributions. As part of this discussion, I will present the findings of our recent observer experiments intended to evaluate the effects of texture orientation, and other characteristics, on shape judgment accuracy. Time permitting, I will also describe some of our recent work in developing methods to more accurately estimate the continuous geometric properties of smooth surfaces approximated by arbirary triangle meshes, and in efficiently synthesizing shape-conforming textures.

CREEDA is an integrated Web-based application for planning agricultural activities. This application incorporates Web versions of such applications as RUSLE (Revised Universal Soil-Loss Equation), STMDL (Sediment Total Maximum Daily Load), WebPST (Web-Based Pesticide Screening Tool), SISL (Surface Irrigation Soil Loss), and SCI (Soil Conditioning Index). Web-based map interfaces are an important feature of these applications. Each application uses a map generated by ArcIMS as the primary user interface. With this map interface, the user can insert, query, update, and delete map features and the information associated with them. A relational database and an ArcSDE server in combination manage the geospatial data. The ArcIMS Internet map server generates maps to be displayed on a Web browser by using the geospatial data provided by the ArcSDE server. The Web pages, including those that display maps, are generated by the server-side scripts written in ASP.NET.

Biography

Measured impedances of mm-wave components in the B.S. project. Designed and implemented a digital control system for a rate-gyroscope in the M.S. project. Enagaged in design and implementation of computer-control systems for fossil-fuel and nuclear power plants for six years. Obtained a Ph. D. degree by working on concurrency and recovery schemes for distributed database systems. Created many sample programs as active object systems. Currently working on Web-based applications that use databases and GIS interfaces. A licensed first-class radio engineer (Japan) and a licensed fisherman (Oregon).

Tools and environments to enable end users to "program" are becoming increasingly popular. The best known such environment is the spreadsheet, and the way users program in this type of environment is by providing formulas. In this talk, we will consider what happens when we add to end-user programming environments consideration of elements of the software engineering lifecycle beyond coding. Doing so seems necessary, because there is ample evidence that end users' programs are no more reliable than those written by professional software engineers. My colleagues and I have been developing a holistic approach to software engineering for end users. It incorporates support for testing, finding bugs, and specification, in an incremental manner integrated in a fine-grained way with the programming environment. The software engineering knowledge needed is in the system, and the user is not expected to develop expertise at software engineering; instead, the strategy is for the system to provide guidance to the user. In the talk, I will focus primarily on how testing and assertions are supported as part of this holistic approach. A number of papers on this topic are available at http://www.cs.orst.edu/~burnett/reprints.html

Biography

Margaret Burnett has been involved in visual programming language research for many years, and most recently her interests have turned especially to end-user programming and end-user software engineering. She is the principal architect of the Forms/3, a spreadsheet-like research language for exploring the boundaries of the spreadsheet paradigm, and of the FAR multi-paradigm programming language for end users. Burnett was recently honored with Oregon State University's Elizabeth P. Ritchie Distinguished Professor Award. She is also a past recipient of the National Science Foundation's Young Investigator Award. She has been a member of the Program Committees for the IEEE Visual Language Symposium, ACM Conference on Programming Language Design and Implementation, and the ACM Conference on Functional Programming, and several others. She has also held various offices in conference committees of the IEEE Visual Language Symposium, ACM Multimedia, and ACM/IEEE International Conference on Software Engineering. For more information about her research and/or papers, see http://www.cs.orst.edu/~burnett/

At the heart of the success of the Internet is its congestion control behavior. TCP is the dominant congestion control protocol in the Internet, and it determines the bandwidth allocation among TCP flows. In recent years, multimedia applications have become increasingly popular in the Internet. TCP, although powerful and effective, is not sufficient to satisfy multimedia application well. Therefore, many congestion control protocols have been proposed recently particularly for streaming media in the Internet. To ensure that flows using a new protocol share bandwidth fairly with TCP flows, we have to predict the bandwidth share ratio among them before deploying new protocols. The prediction can be done in a static way that assumes a constant congestion signal for all traffic. It can also be done in a dynamical way that associates each individual flow's congestion perception to its rate behavior. We choose a dynamic approach and model the bandwidth share among competing traffic with a state-space model. We use the model to describe the stability of bandwidth competitions, which is characterized as convergence to a dynamically oscillating limit cycle in the state space. Real-world experiments confirmed to us that the dynamic modeling produces results that closely match real measurements, while some statically derived share ratios sometimes do not.

Biography

Kang Li is a Postdoc in the System Software Laboratory in the Computer Science and Engineering Department at Oregon Graudate Institute (OGI) in Beaverton, Oregon. He received his Ph.D in Computer Science and Engineering from OGI in October, 2002. His research interests are in the area of computer networks, in particular, congestion control and network measurements.

As Open Source software has matured, so have the tools that enable their development, production, distribution, and use of feedback management. Web-based Portal Communities issued under the General Public License (GPL) that support Open Source software development have sprung up virtually around the world. One significant tools source has been the VASoftware SourceForge.net application. The application integrates Task Management, Bug Tracking, Support Request handling, file distribution, database I/O, and source configuration management (SCM) into a common Web-enabled platform. SourceForge success has been ensured through its use by more than 500,000 software developers working on more than 50,000 projects. Many SourceForge-based communities have been formed beyond the VASoftware sponsored portal. Some communities are open to the public at large. Others, in fact a significant number, have been brought behind corporate firewalls and provide the basic tools needed to efficiently produce software in a closed setting. Expanding SourceForge deployments one step further, Tektronix has made a donation of a fully functioning SourceForge Community to Oregon State University. It is called the Oregon Software Technology Exchange (OSTE). The power and potential of OSTE will be presented and discussed. The impact that OSTE provides by integrating Web-based software development tools will be illustrated.

Biography

Christopher Perez currently program manages initiatives which span Tektronix, Inc. engineering, manufacturing, corporate IT, and customer service functions. He is helping to solve challenges involving globally distributed software design centers, timezones, countries, and cross cultural information exchanges are addressed. He is helping investigate solutions to global software engineering project management challenges which include technical information, software configuration management, defect tracking, production release, and intellectual property distribution, management, and historical archive creation. Christopher's 20 year employment history includes providing products covered under the Open Source Movement, X terminal and Test and Measurement product development, WAN and LAN RF client server technology deployments, and aerospace software design and implementation. The Tektronix Web site is www.tektronix.com

Software as a technology is fairly well understood, but software as a business is not. The relation between the two is so unique, challenging and important that some business schools and computer science departments have begun to address it specifically. This is timely, for while the majority of software professionals continue to work in IT departments of companies whose business is not software, these jobs are increasingly eroded by commercial software and outsourcing. Until it fully embraces the issues of software commercialization, software development cannot achieve relative security as a career or true engineering status as a discipline, nor can it fulfill its inherent potential to build significant personal and economic wealth. Effective strategic planning for today's high-volume software markets requires an understanding of how software market structure evolves, and how these markets move toward value. This requires a new paradigm, or way of seeing software value. The conventional view that software value derives mainly from internalities, such as features or quality, is dangerously blind to the critical externalities that come to dominate actual perceived value in higher volume software products. Successful high-volume software products are designed for total value and are managed as an evolving release portfolio with a compelling theme and trajectory that recruits and orients customer demand. Certain product strategies can reorient demand away from competitors' trajectories, dramatically changing the direction of both the technology and the market. This talk, targeted to both computer science and business professionals, gives a brief overview of modern value driven software commercialization issues and strategies.

Biography

Frank Hall has more than 25 years experience in software development, management and research. He is president of EntreDigm Consulting LLC, a corporate and business development consultancy for software and Internet companies (www.entredigm.com). He is a co-founder and board director of Infinity Softworks, the global market leader in software calculators for handheld computers (www.infinitysw.com), and a former board director of Camo, an international statistical software and services corporation (www.camo.com). He serves as a business development executive for ProWorks, maker of data visualization software (www.proworks.com), and has similarly served Infinity Softworks, Camo and PrintQuick, maker of Internet print delivery software (www.printquick.com). Frank Hall is an adjunct faculty member of OHSU's OGI School of Science and Engineering, where he has taught a graduate class in software commercialization for OHSU and PSU (www.ogi.edu/MST/classes/MST531). He is a board director of the Software Association of Oregon, and serves as president of its Corvallis chapter. Previously he served Hewlett-Packard for 12 years as an R&D project manager developing open systems software, including the Motif user interface toolkit. His research in the mid-1990s into software business success factors led him to leave HP and launch his consulting firm. He has published in IEEE Software and the HP Journal. He completed all but thesis toward an MS in Computer Science at OSU, and holds an MS degree in cultural anthropology from the University of Texas at Austin, and a BA in mathematics from Florida State University. He lives in Corvallis with his wife and two children.

If the level of adoption of Software Engineering Best Practice is to be increased in industry, then it must be taught effectively in the university. Many Computer Science students, however, either view software process as intellectually shallow or are averse to the oppressive discipline which they perceive to be required to follow it. We have devised a method, Live-Thru Case Histories, for motivating students to recognize the necessity of Software Engineering Best Practice, and to learn how to use it, by shocking them into the realization that without it they are likely to fail, not in their course work, but in real-world software development projects. The method has been used for three years as part of a required Senior Project/Software Engineering course, at Stevens Institute; it was recently used in Barry Boehm's graduate Introduction to Software Engineering at USC, and has been used at a number of other universities to good effect. Its effectiveness is being assessed through the use of a number of instruments which we have developed, including an Attitude Toward Software Engineering (ATSE) survey. ATSE has been validated through administration and focus groups at meetings of NJSPIN (North Jersey Software Process Improvement Network), LASPIN, Southern California SPIN, Xerox Corporation's Software Engineering Process Group, and DoD's Software Technology Conference. We discuss the Live-Thru method, the use of ATSE in assessing its effectiveness, and the more general use of ATSE in outcomes assessment of Software Engineering courses, degree programs in Computer Science and Software Engineering and of software process improvement efforts in industry.

Biography

Prof. Klappholz has a BS in mathematics and linguistics from MIT, and an MSEE and PhD in Computer and Information Science from the University of Pennsylvania. He has taught at Columbia University, Polytechnic University, and Stevens Institute of Technology. His technology research areas have included parallel computer architecture, compilation techniques for parallel architectures, and general compiler optimization techniques. He has recently become interested in software development process. His most recent research interests include techniques for measuring attitude toward, knowledge of, and ability to apply, software development process/best practice, pedagogic techniques for teaching process/best practice, empirical software engineering studies, especially within the classroom, and the extension of static, point solution cost/schedule/quality estimation tools to automated tools for continuous estimation and (feedback-based) software project management. He has recently completed a sabbatical at USC, where he worked with Barry Boehm and Dan Port on various software development research and pedagogy issues.

Camera calibration is one of the classical problems in Computer Vision. It consists in estimating the internal geometry of the camera, namely the image formation process through its optical system, as well as the external geometry of the camera, i.e., its location and orientation in space. The pinhole camera is a convenient abstraction for modeling any optical acquisition device: 3-D points are mapped into 2-D image points through a simple perspective projection. However, unlike the ideal pinhole camera, any real camera is equipped with lenses which inevitably introduce image distortion. The estimation of the parameters describing such distortion and the estimation of the actual focal length of the camera provide the complete characterization of the internal geometry of the camera. The knowledge of the external geometry is not necessary in many applications but it has to be obtained anyway since its parameters are intertwined with those of the internal geometry. In this colloquium, I will present a new and very efficient method for the complete calibration of a digital camera.

Biography

Dr. Luca Lucchese received the M.S. degree and the Ph.D. in electrical engineering from the University of Padua, Italy, in 1993 and 1997, respectively. From 1997 to 2002, he was with the Department of Electrical and Computer Engineering of the University of California, Santa Barbara, first a post-doctoral researcher and then as a visiting assistant professor. Since March 2002, he has been with the Department of Electrical and Computer Engineering at Oregon State University as an assistant professor. He teaches courses of digital signal processing and image processing. His research interests include motion analysis and estimation, image registration and mosaicking, three-dimensional imaging, and color image processing. Dr. Lucchese is an Associate Editor of the IEEE Transactions on Image Processing.

The TaskTracer Project is a newly initiated computing research project at Oregon State University that is investigating the possibilities of a desktop software system that will track in detail how knowledge workers complete tasks, and intelligently leverage that information to increase efficiency and productivity. The goal is to create a system with four capabilities: more efficient task-interruption recovery, individual knowledge management, workgroup knowledge management, and within-workgroup workflow detection and analysis. The proposed system will operate in the Microsoft Windows environment, tracking most interactions with desktop applications as well as tracking phone calls. A central challenge involves “unweaving” the threads of events that are affiliated with different tasks that are being performed concurrently. Our approach will combine creative user interfaces and machine learning to perform this unweaving. Other challenges include identifying “significant” events, summarizing tasks or subtasks, predicting likely steps and resources, and detecting cross-user workflow. The TaskTracer project at Oregon State is part of the Management of Knowledge-Intensive Dynamic Systems (MKIDS) initiative funded jointly by the National Science Foundation and the Intelligence Community.

Biography

Jon Herlocker (BS Lewis & Clark College 1994; PhD Minnesota 2000) is Assistant Professor of Computer Science at Oregon State University. Dr. Herlocker's research work centers on integrating intelligence and usability into information systems. At the University of Minnesota, he was part of a group of researchers that developed a collaborative filtering-based recommender system which was licensed to Net Perceptions Inc, and was the lead designer and developer for the MovieLens web-based movie recommender (www.movielens.org). He has published a number of scientific papers on algorithms and systems for collaborative filtering, and has been awarded the prestigious NSF Faculty Early Career Development (CAREER) award for his work in that area. Dr. Herlocker is a member of the ACM and ACM SIGCHI, the IEEE and the Computer Society, and is the faculty advisor the local student chapter of the ACM. http://cs.oregonstate.edu/~herlock/

The schema for a relational database contains key information about the structure of the information it contains. Exploitation of this information can substantially enhance the power and ease-of-use of machine-learning and data-mining systems. In this talk I will review recent and ongoing research in developing a tractable language for specifying derived variables in probabilistic model discovery from relational data sources.

Biography

Dr. D'Ambrosio is an Associate Professor in Computer Science at Oregon State University. He is also founder of CleverSet, Inc. His research focuses on representation, inference, and discovery of relational probabilistic models, especially for such tasks as situation assessment and process modeling (e.g., web user behavior, west-nile virus spread)

http://www.cs.orst.edu/~dambrosi

The People and Practices Research Lab at Intel consists of about a dozen social scientists, designers, and engineers. We study real people across the globe in their natural work and life environments in order to imagine new uses for computing power, to identify unmet needs and desires, and to understand barriers to technology adoption. In this talk, I will introduce the methods and perspective of the group, using examples from recent work in US living rooms, Korean homes and "PC rooms", and Chinese workplaces.

Biography

Scott Mainwaring joined Intel's People and Practices Research Lab in 2000. His research interests include: community, trust, and the social use of technologies; new technologies for homes and families; and the influence of geography and culture on technology adoption and use. Prior to joining Intel, he was a researcher at Interval Research Corp. in Palo Alto, CA for six years, conducting ethnographic fieldwork and prototyping studies in order to better understand the real and potential roles of technology in everyday life. In earlier lives, he could be found developing Unix applications, administering an undergraduate cognitive science program, or studying the formation of spatial mental models in college sophomores. Scott holds an A.B. in computer science from Harvard University and a Ph.D. in cognitive psychology from Stanford University. (See http://www.intel.com/research/people/mainwaring_s.htm)

Most cost-effective digital cameras use a single image sensor, applying alternating patterns of red, green, and blue color filters to each pixel location. A way to reconstruct a full three-color representation of color images by estimating the missing pixel components in each color plane is called a demosaicing algorithm. This paper presents three inherent problems often associated with demosaicing algorithms that incorporate directional interpolation: misguidance color artifacts, interpolation color artifacts, and aliasing. The level of misguidance color artifacts present in two images can be compared using metric neighborhood modeling. The proposed demosaicing algorithm estimates missing pixels by interpolating in the direction with fewer color artifacts. The aliasing problem is addressed by applying filterbank techniques to directional interpolation. The interpolation artifacts are reduced using a nonlinear iterative procedure. Experimental results using digital images confirm the effectiveness of this approach.

Biography

Keigo Hirakawa received the B.S. degree in electrical engineering from Princeton University, Princeton, NJ, in 2000. He is currently pursuing the MS/Ph.D. degree at Cornell University, Ithaca, NY. His research interests include image modeling, color representation, multi-rate systems, and image interpolation. He also pursues a professional career as a jazz pianist.

In a balanced code each code word contains equal number of 1's and 0's. These codes find applications in many areas; they can be used to detect unidirectional errors in VLSI systems, to design fault tolerant and fail-safe sequential circuits, to achieve data integrity in write-once memories, as line codes in fiber optic data transmission systems, as modulation codes in optical and magnetic storage systems, for high speed VLSI design, etc. In coding theory, efficient design of encoding and decoding of balanced codes (i.e. converting data words to balanced codes and from balanced words to data words) has been an open research problem for many years. In 1986, Knuth has given some simple design schemes for these codes. Since then we have given many improved design methods. In this talk, after describing some of the applications, some design methods will be described.

Biography

http://cs.oregonstate.edu/~bose/cv.html

Walden's Paths enables the specification and viewing of presentations formed out of materials gathered from throughout the World-Wide Web. This is accomplished by layering a metastructure--in this case a linear path--on top of already existing materials. The organization of materials into paths allows (indeed requires) the inclusion of annotation in order to provide contextualization to the reader. Ownership of Web materials is administratively decentralized, and this raises challenges for Path maintenance; essentially, a goal of Walden's Paths is to build a stable structure on a constantly shifting substrate. The talk will describe Walden's Paths and its Path maintenance subsystem. Further information about the project is located at http://www.csdl.tamu.edu/walden.

Biography

Richard Furuta is a faculty member at Texas A&M University where he is a Professor in the Department of Computer Science, Director of the Hypermedia Research Laboratory, and Associate Director of the Center for the Study of Digital Libraries.

Dr. Furuta's current areas of research include digital libraries, hypermedia systems and models, structured documents, and document engineering. He also has studied applications in computer supported cooperative work, software engineering, visual programming, document structure recognition from bitmapped sources, and management systems for three-dimensional-gesture-based user interfaces. In the area of Digital Libraries, he was one of the founders of the 1994 and 1995 Digital Libraries Conferences, which subsequently became the ACM Digital Libraries series, and later merged with the IEEE-CS series to form the ACM/IEEE-CS Joint Conference on Digital Libraries (JCDL). He was program chair for ACM Digital Libraries 2000, currently serves as Chair of the Steering Committee for ACM/IEEE-CS JCDL, and as an Editor-in-Chief of the Journal of Digital Libraries. In other technical activities, he has been co-program chair for ACM Document Engineering 2002, co-program chair for ACM Hypertext '93, program chair for Electronic Publishing '90, co-program chair of the 1991 DC ACM Chapter annual symposium, Chair of ACM SIGLINK from 1993-1995, member of the ACM SIG Board/SIG Governing Board Executive Committee from 1997-2001, and has served on many other program committees, conference committees, steering committees, and editorial boards.

NSF's latest "Major Research Equipment" project intends to revolutionize how earthquake engineering research is conducted. What will make this possible is an ambitious infrastructure revolving around advanced information technology: very highspeed networking, collaboration technologies, distributed data acquisition, large-scale data storage and management, high-performance computing, and augmented reality. Its success will depend not just on access to these special resources, but on how usable we can make them for engineering researchers. This presentation describes how advances in computer science are being harnessed to change the nature of engineering research. It focuses on the usability challenges that must be addressed so that a national community of researchers can share state-of-the-art laboratory facilities, data, computational models, and research experiences. Examples are shown of what OSU computer scientists are contributing to this national effort.

Biography

Cherri M. Pancake is Professor of Computer Science and Intel Faculty Fellow at Oregon State University. Her previous career involved extensive ethnographic fieldwork, where she applied cross-cultural survey and interviewing techniques to study social change in Guatemalan Indian communities. After earning a Ph.D. in Computer Engineering from Auburn University, Pancake began applying both ethnographic and engineering techniques to the problem of how software tools and Web-based interfaces can more closely match users' intuitions. Pancake is director of the Northwest Alliance for Computational Science and chair of the Parallel Tools Consortium, both collaborative efforts involving computer scientists and scientists from a wide variety of disciplines. She serves as Strategic Advisor on Usability for the San Diego Supercomputer Center, as well as leading the information technology portion of the consortium developing NSF's George E. Brown Network for Earthquake Engineering Simulation. Pancake is a Fellow of both the ACM and the IEEE. She is also currently serving as department head of Computer Science. See www.cs.orst.edu/~pancake and www.nacse.org.

Computational and data Grids couple geographically distributed resources such as high performance computers, workstations, clusters, and scientific instruments. Accordingly, they have been proposed as the next generation computing platform for solving large-scale problems in science, engineering, and commerce.

To date there are very few examples of programming environments that allow legacy applications to be “Grid Enabled”, and thus all Grid demonstrators have been constructed from scratch. Middleware software layers like Globus and Legion are powerful, but they tend to provide a set of low-level primitives which must be called from within the application. This means that at present, in order to build a general Grid application, it is necessary to modify legacy code.

In this seminar I will present two projects in which the goal has been to Grid enable legacy software. The Nimrod project http://www.csse.monash.edu.au/~davida/nimrod.html) has targeted parameter sweep studies in which many independent tasks are distributed to grid resources. Nimrod uses a novel computational economy to enforce a deadline based quality of service system. The GirddLeS project (http://www.csse.monash.edu.au/~davida/griddles.html) targets more general grid applications built from a set of co-operating legacy components. These are linked by a communication protocol called GridFiles.

The architectural design of a software system can support analysis of high-level properties and provide benefits for many software evolution tasks. However, existing tools decouple the architecture of a program from its implementation, allowing inconsistencies to accumulate as the system evolves. Because of the potential for inconsistency, engineers evolving a program cannot trust the architecture to accurately describe the properties or structure of the implementation. This talk presents a new approach: integrating architectural descriptions into an implementation language, and using a type system to ensure that the architecture is consistent with the code. The approach is embodied in the ArchJava language, which extends Java with features that declare the software architecture and data sharing within a system. ArchJava's type system enforces architectural conformance, the property that implementation components communicate only in the ways specified by the architecture. ArchJava is flexible enough to describe hierarchical architectures that may change at run time, and it supports many of the same coding styles and idioms that programmers use in Java. Several case studies applying ArchJava to existing programs provide preliminary evidence that ArchJava is practical and can aid software evolution tasks.

The new OSU Online Catalog & Schedule of Classes was developed using Microsoft ASP.NET and ADO.NET technologies. We will take a look at how these technologies, along with the Object Oriented nature of the .NET Framework, accelerated the development, and facilitated prototyping used to stimulate user feedback during the development process. Some of the key aspects of the architecture of the application will also be discussed. The Online Catalog & Schedule of Classes can be found at http://catalog.oregonstate.edu.

Biography

Mark Clements graduated from Oregon State University with a BA in Spanish in 2001. During his years as a student he worked as a computer lab manager for the College of Business and worked his way into web development there. He is now a Senior Systems Development Engineer for the Business Solutions Group in the College of Business, of which he is a founding member. Mark is a self-taught developer and has been developing .NET applications and services for 2 years.

Bayesian belief networks are being implemented with increasing frequency in ecological modeling. In particular, ecologists are interested in understanding the mechanisms by which stressors to the environment affect the health of ecosystems. Current methodology for fitting Bayes networks involves an assumption of spatial independence that is improbable in ecological settings. In this talk, I will describe some relatively simple Bayesian networks for modeling an ecological system, and describe some of the statistical issues for fitting those networks.

Biography

Dr. Gitelman has been at OSU for 4 years and has a doctorate in Statistics from Carnegie Mellon University. In her thesis research she extended a class of causal models to the setting of hierarchical models. Currently, Dr. Gitelman is working in Bayes networks, which are another class of causal model, but not unrelated to hierarchical models.

Tom Dietterich and Adam Ashenfelter
OSU Computer Science

Recently, Lafferty, McCallum and Pereira introduced the Conditional Random Field as a new model for solving sequential supervised learning problems. Many applications of machine learning can be formalized as Sequential Supervised Learning (SSL). Each training example in SSL has the form (X,Y), where X is a sequence (x1, ..., xT) of items (each typically described by a vector in R^n), and Y is a sequence (y1, ..., yT) of class labels from {1, ..., K}. Typical applications include part-of-speech tagging, information extraction from web pages, and text-to-speech mapping. There are no robust, off-the-shelf methods for solving SSL problems in any commercial or academic statistical or data mining software systems. This talk will provide an introduction to the problem, discuss why the CRF is a good candidate for an off-the-shelf method, and describe our work on applying Friedman's Gradient Tree Boosting algorithm to efficiently and flexibly fit CRF models to the large NETtalk text-to-speech data set. Preliminary performance results will be presented.

Biography

http://www.cs.orst.edu/~tgd

Dr. Dietterich (AB Oberlin College 1977; MS University of Illinois 1979; PhD Stanford University 1984) joined the OSU faculty in January 1985. In 1987, he was named a Presidential Young Investigator for the NSF. In 1990, he published, with Dr. Jude Shavlik, the book entitled Readings in Machine Learning, and he also served as the Technical Program Co-Chair of the National Conference on Artificial Intelligence (AAAI-90). From 1992-1998 he held the position of Executive Editor of the journal Machine Learning. The American Association for Artificial Intelligence named him a Fellow in 1994, and the Association for Computing Machinery did the same in 2003. In 2000, he co-founded a new, free electronic journal: The Journal of Machine Learning Research. He served as Technical Program Chair of the Neural Information Processing Systems (NIPS) conference in 2000 and General Chair in 2001. He currently President of the International Machine Learning Society and he also serves on the Board of Trustees of the NIPS Foundation.

People increasingly use the web not only to browse information, but to perform actions, such as online purchases. Existing web browsers do not provide specific support for recording and reviewing these online actions, however. This support is particularly necessary when something goes wrong and users need to debug the processes they initiate. I will present Woodstein, an agent that monitors and displays user actions on the web. Woodstein explains the details as well as the structure of these processes and also supports debugging by helping users keep track of their own debugging process.

Biography

Earl Wagner is a masters' student at the MIT Media Laboratory. He is interested in software that helps users understand and modify the systems they interact with, especially when something goes wrong. He is currently working with Henry Lieberman to develop Woodstein, an end-user debugger for e-commerce that helps users understand their actions on the web. Before coming the Media Lab, he researched technologies for software development environments and received a B.S. in computer science from the University of California at Berkeley. Homepage web.media.mit.edu/~ewagner

Biography

Rick Lindsley is a software engineer with 20 years experience in Unix and Linux. He's worked on everything from Berkeley 4.2 and 4.3 releases to Solaris to, now, Linux. He spent several years at Tektronix in Beaverton, Oregon, then gained significant experience with multi-processor machines while working for Sequent Computer Systems for nine years. When Sequent was purchased by IBM in 1999, he stayed on board and joined IBM's Linux team. He's currently working in the Linux Technology Center for IBM in Beaverton Oregon. Since then he's given talks at universities, national laboratories, and Linux user groups on topics ranging from what it's like to work for IBM, to issues with SMP locking, to the current state of the Linux kernel. His recent focus is on I/O statistics gathering and process scheduling improvements.

Multithreaded architectures have received considerable attention over the past few years. However, these architectures rely on conventional programming paradigms and require complex runtime transformation of the control-flow programs into dataflow programs, requiring complex hardware to detect data and control hazards, reorder and issue multiple instructions. Our architecture differs from other multithreaded architectures in two ways: i) our programming paradigm is based on dataflow, which eliminates the need for complex runtime scheduling, thus reducing the hardware complexity significantly, and ii) complete decoupling of all memory accesses from execution pipeline. The underlying dataflow and non-blocking models of execution permit a clean separation of memory accesses (which is very difficult to coordinate in other programming models). Data is pre-loaded into an enabled thread's register context prior to its scheduling on the execution pipeline. After a thread completes execution, the results are post-stored from its registers into memory. The instruction set implements dataflow computational model, while the execution engine relies on control-flow like sequencing of instructions. Unlike Superscalars, our architecture performs no (dynamic) Out-of-Order execution and thus eliminates the need for complex instruction scheduling hardware. In this talk I will present the performance comparisons of our architecture with superscalar and VLIW systems for various benchmarks. The results show that our architecture scales better than other systems when more resources are added.

Biography

Krishna Kavi is currently a Professor and the Chair of Computer Science and Engineering department at the University of North Texas, in Denton, Texas. Prior to joining UNT, he held faculty positions at the University of Alabama in Huntsville and the University of Texas at Arlington. He also served as a Program Manager at the National Science Foundation. He was an IEEE Computer Society (CS) Distinguished Visitor (1989-91), editor of the IEEE Transactions on Computers (1993-1997), and editor of the Computer Society Press (1987-1991). His primary research interest lies in Computer Systems Architecture, including dataflow and multithreaded systems, Memory Management, Operating Systems, and Compiler Optimization. His other research interests include Formal specification of concurrent processing systems, Performance Modeling and Evaluation, load balancing and scheduling of parallel programs. He published over 125 technical papers on these topics. He received his B.E. (Electrical) from the Indian Institute of Science, MS and Ph.D. (Computer Science) from the Southern Methodist University.

With the increasing availability of genomics and proteomics tools that identify new genes, proteins, and biological pathways in which they take part, the ability for biologists to move into new areas of the scientific literature and other resources - a.k.a., the bibliome - is essential. Although the bibliome is widely available through the Internet, considerable work is still required to find and synthesize available information. The Text Retrieval Conference (TREC) is an annual forum for researchers in information retrieval (IR) to evaluate their systems and approaches with a common data set. Most work in TREC has focused on the newswire domain, but a convergence of researchers from the biological and IR fields who are interested in enhancing access to information in the genomics domain has emerged. Dr. Hersh, a noted authority in IR in the biomedical domain will describe the new TREC Genomics Track which aims to advance IR systems for genomics researchers and in turn allow them to improve their research.

Biography

Rick Lindsley is a software engineer with 20 years experience in Unix and Linux. He's worked on everything from Berkeley 4.2 and 4.3 releases to Solaris to, now, Linux. He spent several years at Tektronix in Beaverton, Oregon, then gained significant experience with multi-processor machines while working for Sequent Computer Systems for nine years. When Sequent was purchased by IBM in 1999, he stayed on board and joined IBM's Linux team. He's currently working in the Linux Technology Center for IBM in Beaverton Oregon. Since then he's given talks at universities, national laboratories, and Linux user groups on topics ranging from what it's like to work for IBM, to issues with SMP locking, to the current state of the Linux kernel. His recent focus is on I/O statistics gathering and process scheduling improvements.

The advent of multi-agent systems (MAS) has brought us opportunities for the development of complex software that will serve as the infrastructure for advanced distributed applications. During the past decade, there have been many agent architectures proposed for implementing agent-based systems, and also a few efforts to formally specify agent behaviors. However, research on narrowing the gap between agent formal models and agent implementation is rare. In this talk, we present a model-based approach to designing and implementing multi-agent software systems. Instead of using formal methods for the purpose of specifying agent behavior, we bring formal methods into the design phase of the agent development life cycle. During the presentation, we first introduce the formalism called agent-oriented G-net model, which is based on the G-net formalism (a type of Petri nets), to serve as the high-level design for intelligent agents. To illustrate our formal modeling technique for multi-agent systems, an example of an agent family in electronic commerce is provided. Then we show how an existing Petri net tool can be used to detect design errors, and how model checking techniques can support the verification of some key behavioral properties of our agent models. Finally, based on the high-level design, we derive the agent architecture and the detailed design for agent implementation. To demonstrate the feasibility of our approach, we developed the toolkit called ADK (Agent Development Kit) that supports rapid development of application-specific agents for multi-agent software systems.

Related URL: http://www2.uic.edu/~hxu2/PhD/Thesis.html

Biography

Haiping Xu received the BS degree (1989) and the MS degree (1992) in electrical engineering from Zhejiang University, Hangzhou, China, the MS degree (1998) in computer science from Wright State University, Dayton OH, and the Ph.D. degree (2003) in computer science from the University of Illinois at Chicago, IL. From 1992 to 1996, he successively worked with the Ministry of Electronics Industry, Shen-Yan systems Technology, Inc. and Hewlett-Packard Co., as a software engineer, in Beijing, China. He was a research scholar at Nanyang Technological University, Singapore, during the summer of 1996. His research interests include distributed software engineering, multi-agent systems and model-based software development. He is a member of the ACM, IEEE and IEEE Computer Society.

URL: http://www.cs.uic.edu/~hxu1

I explore problems with many "maybes". In experiments with randomized world generation from abductive inference engines, I saw that (often) a few randomly selected worlds of belief yielded as much useful information as searching many more worlds.

This result seemed crazy- a few quick peeks are as good as many hard stares? Yet after much experimentation, I can report that the effect is repeatable in many domains.

After many years exploring this effect this talk can report WHEN we can expect the effect to repeat, WHY the effect happens, and HOW we can use it to dramatically simplify software engineering.

Biography

Dr. Tim Menzies has been working on advanced software engineering techniques since 1986. He is the author of over 150 research papers and currently is the software engineering research chair at NASA's Independent Verification and Validation (IV&V) Facility. Dr. Menzies received his PhD from the University of New South Wales, Sydney, Australia and is a member of the ACM and IEEE.

For more information, see

http://menzies.us
http://menzies.us/me.html
http://menzies.us/papers.html
http://menzies.us/Draftpub.html

Businesses that rely on web services are vulnerable to the problems of those web services. Service contracts and warranties can provide some assurances. However, they provide traditional recourse, rather than timely alerts of impending problems. While electronic commerce has increased the speed of on-line transactions, the technology of monitoring on-line transactions has lagged behind.

To address the problem of web service monitoring, we integrated methods of requirements analysis and software execution monitoring. The resulting system assists analysts in the development of web service requirements monitors.

The work presented here builds on prior research by: (1) building on a goal-based method for obstacle discovery, (2) illustrating the derivation of assigned monitors from obstacles, and (3) automatically deriving web service monitors from high-level requirements descriptions. The framework, and tool, provides an important contribution by demonstrating how distributed concurrent web service transactions can be monitored at the requirements level.

Biography

Dr. William Robinson is an associate professor at Oregon State University (COB/MIS). He has written over 40 academic articles, mostly in the areas of Requirements Engineering and agent support of Electronic Commerce. His journal articles include Communications of the ACM, IEEE Transactions on Software Engineering, Journal of Global Information Management, Concurrent Engineering: Research & Applications, and ACM Computing Surveys. Dr Robinson is secretary of IFIP Working Group 2.9 (Software Requirements Engineering). He is on the steering committee of the IEEE Requirements Engineering Conference series, and was past program chair of the Fourth IEEE Requirements Engineering Symposium (RE'99).

As CPU speeds and counts have grown in recent years, the cost of maintaining symmetric access across the system has greatly increased. To alleviate these costs machines with Non-Uniform Memory Access (NUMA) have appeared.

The 2.6 Linux kernel includes many enhancements in support of NUMA machines. Data structures and macros are provided within the kernel for determining the layout of the memory and processors on the system. These enable the memory subsystem to make decisions on the optimal placement of memory for processes, and the scheduler to factor in node locality when making process scheduling and load balancing decisions. In addition to items that have been incorported into the mainline Linux kernel, there are NUMA features that have been developed that continue to be supported as patchsets.

This talk will cover 2 specific features that have grown out of these development efforts: kernel text replication and user page replication. Both of these features aim to make local copies of read-only data to enhance NUMA performance, but the solutions are radically different.

Biography

Dave Hansen graduated from Purdue University with a B.S. in Computer Science in 2001. He joined IBM's Linux Technology Center in Beaverton, Oregon where he works with the Linux community to increase kernel scalability. His work has run the gamut from SMP locking, to filesystems, to networking, but his current work is on the Linux virutal memory subsystem.

First and foremost, we all [should] know that the real world is continuous and ANALOG (unless you dig deep down to the discrete nature of photons). Despite the fast growth of digital systems today, an analog circuit such as analog-to-digital (A/D) converter is a necessary component in just about all digital systems, as the real-world interface between analog and digital has to be addressed at some point. In the area of analog chip design, the shrinking transistor dimensions (which is good for digital like the pentium processor) is anticipated to create some big problems. This is because small transistors can tolerate only a small amount of voltage stress. A 0.18 micron CMOS process, for example, can only tolerate 1.8 volts. My seminar will briefly review this low-voltage problem; summarize some of the well-known solutions currently in use (and problems associated with these solutions); and new circuit techniques that we have recently developed here at Oregon State University. Some chip implementation results of pipelined A/D converter will be presented. I will also devote about 5-10 minutes of my talk to go over my background and some other research topics at OSU that my research group is currently involved in.

Biography

Prof. Moon received B.S. from University of Washington, M.Eng. from Cornell University, and Ph.D. from University of Illinois, Urbana-Champaign. From February 1994 to January 1998, he was with Bell Laboratories. Since January 1998, he has been with Oregon State University. His research area is in analog and mixed analog-digital integrated circuits.

Energy Systems encompasses the areas of Power Electronics, Machines and Drives in Industrial Processes, and Power Systems including Alternative Energy Interfaces. This seminar will present an overview of the Energy Systems research at OSU, which has the central theme of "Improved Power Processing", with a wide variety of interesting projects and applications. Specifically, my research expertise has focused on Power Electronics (power supply design, converters and multi-level inverters), Power Quality and Renewables (including utility interface issues and filtering/conditioning approaches) and Adjustable Speed Drives (application issues including overvoltage, bearing currents and electromagnetic interference, as well as ride-through). This seminar will also describe the Motor Systems Resource Facility (MSRF) - the highest power Energy Systems laboratory in any university in North America (including a 750kVA dedicated power supply, comprehensive testbeds up to 300hp, and a 120kVA fully programmable source). www.ece.orst.edu/~avj, eecs.orst.edu/msrf/

Biography

Dr. von Jouanne is in her 9th year as a professor in the School of EECS at OSU. She is in the Energy Systems group working primarily on power electronic converters and industrial drives, power quality and renewables. She is the Co-Director of the Motor Systems Resource Facility, a registered professional engineer, and was the recipient of the 2000 IAS Outstanding Young Member Award. Dr. von Jouanne received her Ph.D. degree in Electrical Engineering from Texas A&M University where she also worked with Toshiba International Industrial Division. She received her M.S. and B.S. degrees in Electrical Engineering with a Minor in Mathematics from Southern Illinois University. She has been interviewed for the internationally-syndicated NPR program "51%", highlighting women who have impacted society, especially through science and technology. She has also been selected by the National Academy of Engineering to be profiled as one of their "Celebrated Women Engineers". www.ece.orst.edu/~avj

A brief overview of ongoing Materials & Device research at OSU is presented. This work is an interdisciplinary effort involving researchers in the Departments of Chemistry, Chemical Engineering, EECS, and Physics. Research topics to be surveyed include transparent electronics, low-cost electronics, photovoltaics, and intelligent luminescence.

Biography

J. F. Wager is a Professor in the School of EECS at OSU whose research specialization is in the area of solid-state materials and devices.

http://eecs.oregonstate.edu/research/members/wager/index.html

Reinforcement Learning (RL) is the study of systems that improve their performance at some task by taking actions and receiving rewards and punishments. The rewards are propagated backwards in time so that the system prefers actions that lead to large rewards in the long run. Reinforcement Learning is an active area of research with many theoretical and practical results. In this talk, I focus on scaling RL to domains with large state spaces through abstraction hierarchies and function approximation. I'll also outline some ideas on scaling RL to large action spaces and to relational domains.

Biography

Prasad Tadepalli has an M.Tech from Indian Institute of Technology, Madras, and a Ph.D from Rutgers University, both in Computer Science. He has been teaching at Oregon State University since 1989. His main area of research is Machine Learning, including Reinforcement Learning, Relational Learning, and Computational Learning Theory with applications to classification, real-time scheduling, and information extraction. http://web.engr.oregonstate.edu/~tadepall

The computing landscape is undergoing a dramatic transformation as computing devices become more ubiquitous, network-enabled and ready to communicate. This trend is not only impacting the way we use computers, but also the way we build software for them. This talk centers on adaptive approaches to distributing content and to performing distributed computations. We first motivate adaptive computing by presenting “Motion,” a mass video distribution network that uses adaptive techniques to deliver video daily to two million users out of “recycled” bandwidth. This approach improves the user experience and significantly reduces the cost of delivery. We’ll then focus on tuple-space models as a basis for adaptive computation; our early work with the Linda coordination language at Yale University suggests that these models have the right characteristics for adaptive computation, while later work with Sun Microsystems on JavaSpaces suggests their utility in a spontaneous network environment. We’ll also introduce and discuss the role of distributed data structures in these computations and their growing presence in today’s Internet. Finally, we will briefly touch on Lifestreams, an interface for the network environment (the speaker’s thesis work), and then suggest future directions in adaptive computing based on current Internet infrastructure and trends.

Biography

Dr. Eric Freeman is currently Director of Engineering for the Walt Disney Internet Group, where he directs broadband and wireless strategy, technology, and product development. Prior to joining Disney, Eric co-founded Mirror Worlds Technologies, Inc., a company that produces a commercialized version of his Ph.D. work. Eric was also previously on faculty at Yale University as a Research Affiliate, where he worked closely with Sun Microsystems and focused on distributed computing using Java and Jini technologies. As a result of this work, Eric co-authored JavaSpaces Principles, Patterns and Practice, the official Sun book on JavaSpaces. Eric received a Ph.D. from Yale University in 1997 for his work on the Lifestreams system, one of the first research systems to suggest an alternative for the desktop metaphor. He was recognized for this work in 1999 by MIT\'s Technology Review as one of the top 100 young innovators.

Eric’s research interests include distributed computing, Internet technologies, information systems and programming languages.

The TaskTracer Project is a newly initiated computing research project at Oregon State University that is investigating the possibilities of a desktop software system that will track in detail how knowledge workers complete tasks, and intelligently leverage that information to increase efficiency and productivity. The goal is to create a system with four capabilities: more efficient task-interruption recovery, individual knowledge management, workgroup knowledge management, and within-workgroup workflow detection and analysis. The proposed system will operate in the Microsoft Windows environment, tracking most interactions with desktop applications as well as tracking phone calls. A central challenge involves “unweaving” the threads of events that are affiliated with different tasks that are being performed concurrently. Our approach will combine creative user interfaces and machine learning to perform this unweaving. Other challenges include identifying “significant” events, summarizing tasks or subtasks, predicting likely steps and resources, and detecting cross-user workflow. The TaskTracer project at Oregon State is part of the Management of Knowledge-Intensive Dynamic Systems (MKIDS) initiative funded jointly by the National Science Foundation and the Intelligence Community.

Biography

Jon Herlocker (BS Lewis & Clark College 1994; PhD Minnesota 2000) is Assistant Professor of Computer Science at Oregon State University. Dr. Herlocker's research work centers on integrating intelligence and usability into information systems. At the University of Minnesota, he was part of a group of researchers that developed a collaborative filtering-based recommender system which was licensed to Net Perceptions Inc, and was the lead designer and developer for the MovieLens web-based movie recommender (www.movielens.org). He has published a number of scientific papers on algorithms and systems for collaborative filtering, and has been awarded the prestigious NSF Faculty Early Career Development (CAREER) award for his work in that area. Dr. Herlocker is a member of the ACM and ACM SIGCHI, the IEEE and the Computer Society, and is the faculty advisor the local student chapter of the ACM. http://cs.oregonstate.edu/~herlock/

Through a series of examples, I will demonstrate presentation techniques that enhance or detract from the message you are trying to deliver. I will discuss how much information to put on a single slide, how slides should relate to one another, subliminal messages slides may convey, when to be silent, the use of gestures, question-answering strategies, and more.

Biography

Professor Michael J. Quinn earned an honors B.S. in mathematics from Gonzaga University in 1977, an M.S. in computer sciences from the University of Wisconsin-Madison in 1979, and a Ph.D. in computer science from Washington State University in 1983. From 1979 to 1981 he worked for Tektronix, Inc. as a software engineer. He was an assistant professor of computer science at the University of New Hampshire from 1983 to 1989, before joining Oregon State University in 1989. He served as interim Head of Computer Science in 1997-1998 and Head of Computer Science from 1998-2002.

Dr. Quinn is the author of dozens of refereed publications in the area of parallel computing. He has also authored or co-authored three books on parallel computing. His fourth book, Parallel Programming in C with MPI and OpenMP, will be published by McGraw-Hill in June 2003. Dr. Quinn is a member of the ACM, the IEEE, the IEEE Computer Society, and Computer Professionals for Social Responsibility. He has received the "Golden Core" service award from the IEEE Computer Society.

The business group SMS (for Secure Mobile Solutions) of Infineon Technologies AG is the fusion of the former business groups WS (for Wireless Solutions) and CC (for Security and ChipCard ICs). I will shortly glimpse over their activities and their products concentrating on those inherently related with security. After a short defense on the importance of merging mobility and security in todays days, I will present the engineering difficulties when developing real-world security products. In addition to that, I will also show how to overcome those security challenges with novel ideas and techniques. All in all, I will guide the audience through the development of a modern high-security micro-controller as used in high-end ChipCard ICs or government crypto-chips.

Biography

Jean-Pierre Seifert received the diploma degree in Mathematics and Computer Science from the Johann-Wolfgang Goethe University at Frankfurt on the Main in 1995. In 1999 he received the Doctorate degree in Mathematics also from the Johann-Wolfgang Goethe University. During 1995 and 2000 he had several visiting positions at the MIT, ETH Zuerich and the Queensland University of Technology. In 2000, he joined Infineon Technologies AG. In 2002 he received Infineon's Inventors of the year" Award for his outstanding numbers of filed patents. Currently he is a Principal Security Research Scientist within the business group Secure Mobile Solutions (SMS) of Infineon Technologies. There, he is leading the "Security Concepts and Innovations" group for the whole SMS business group.

In 2003 he received an offer as a Full Tenure Professor chairing the research field Cryptographic Engineering within the Computer Science Department at the Universitaet Saraviensis (Saarbruecken, Germany). In addition to his large number of patents, he has published several papers on Algorithmic Number Theory, Approximation of NP-hard Problems, Complexity Theory, Computer Arithmetic, Computer Security, Cryptography, Digital Circuit Design, Hardware Design, Lattices in Cryptography and Cryptanalysis, Side channel Analysis, Tamperproof hardware & software design, and also on Quantum Computing.

Many will agree that the fork is perhaps the ultimate recourse when developing extensions to open source and free software projects. Given turnaround time of integration with many large projects, many developers resort to maintaining what are effectively their own forks.

The GUFT Unified Forking Toolkit aims to ease the burden of maintaining common "splinter forks" and aims to make "separate forks" more prevalent. The project's ultimate goal will be to someday see a "Fork Our Distribution" link available from every open source operating system vendor website.

The work will be accomplished with the following set of integrated toolkits:

* GOLT Open-Lean Toolkit (Software Engineering)
* GURT Unified Repository Toolkit (Version Control)
* ARMPIT Recursive Meta-Package Integration Toolkit (Package Management)
* BUZCUT Utterly Zero Click Upgrade Toolkit (Site Management)

Together, these toolkits will provide a comprehensive solution for creating, maintaining, and redistributing splinter or separate forks of complex software projects. In the end, these should allow a single individual or small team to manage several moderately unique distributions. With such a toolkit in hand, companies will be able to leverage forking in tandem with a diversified branding strategy, resulting in what can only be described as "software development franchises".

This talk will explain the rational for creating such a toolkit along with the requirements used to develop the GUFT prototype and its underlying architecture.

Biography

Zach Welch formed Superlucidity Services, LLC to provide consulting services for open source software. After porting Gentoo Linux to ARM, he formed The Zynot Foundation (www.zynot.org) which helped him develop ideas for beneficial forking. He is currently developing a prototype of GUFT and evangelizing the OSU Open Source Lab (www.osuosl.org).

Humanoid robots and agents are becoming increasingly prevalent in a wide variety of applications, ranging from animated human characters to physically embodied humanoid robots. Such humanoids will eventually act autonomously, interacting with humans through collaboration rather than explicit programming. Manually designing and implementing control mechanisms for autonomous humanoids can be a complicated and time-consuming process that lacks scalability and is not suitable for non-technical developers. The aim of my research is to address this problem by automatically designing and implementing modular humanoid skills derived from abilities demonstrated by human beings. In this talk, I describe data-driven methods for unsupervised learning of such skills from human motion.

These methods address two main problems: 1) how can human motion data be analyzed to produce modular humanoid skills and 2) how can suitable human motion data for data-driven analysis be captured. Each of these projects utilizes or extends multidimensional scaling techniques to uncover structure within data produced by human motion in a model-free fashion.

The primary theme of my research is to derive humanoid skills from motion data of humans behaving ``naturally''. An underlying philosophy of this work is that motion capture should require little or no instrumentation such that a subject can move in an unrestricted manner while performing tasks. Towards this end, this talk will describe a method for Kinematic Model and Motion Capture (KMMC). KMMC automatically estimates the kinematic model and joint angle motion from video sequences of multiple calibrated cameras. Furthermore, KMMC can be applied to subjects with arbitrary genus 0 (tree-structured) kinematics. The KMMC method is complimented by Performance-Derived Behavior Vocabularies (PDBV) for deriving ``behavior vocabularies'' from unlabelled motion data of a human performing multiple activities. PDBV uses a spatio-temporal method for multidimensional scaling to uncover behaviors that underlie motion data. Uncovered behaviors can then be realized as nonlinear dynamical systems in the joint angle space of the humanoid. Such dynamical systems are used as predictors for a variety of applications. This talk will demonstrate the utility of behavior vocabularies for humanoid motion synthesis, classification, and imitation.

Biography

Chad Jenkins recently completed his Ph.D. in the Computer Science Department and the Center for Robotics and Embedded Systems at the University of Southern California, under the supervision of Prof. Maja Mataric'. He earned his B.S. in Computer Science and Mathematics at Alma College (1996) and M.S. in Computer Science at Georgia Tech (1998). His research interests include humanoid robotics, machine learning, computer animation, computer vision, and autonomous agents. Recently, Chad presented his research at the International Robots and Systems Conference (IROS), Autonomous Agents and Multi-Agent Systems (AAMAS), and Computer Vision and Pattern Recognition (CVPR).

Low-jitter clock generation and clock/data recovery (CDR) blocks are important components required for reliable operation of high-speed synchronous systems such as microprocessors and data communication links. These blocks must contend with several noise sources that can adversely impact performance. This talk presents a mixed PLL/DLL architecture for low-jitter clock generation and to be used as a core component for clock/data recovery. This architecture merges the characteristics of PLLs and DLLs via an interpolator in order to easily adjust loop dynamics in response to different noise environments.

Biography

Gu-Yeon Wei joined Harvard University in January 2002 as an Assistant Professor of Electrical Engineering in the Division of Engineering and Applied Sciences. Prior to joing Harvard, he spent 18 months at Accelerant Networks in Beaverton, Oregon as a Senior Design Engineer. Professor Wei received his BS, MS, and Ph.D. degrees all from Stanford University in 1994, 1997, and 2001. His current research interests are in the area of mixed-signal VLSI circuits and systems design for high-speed/low-power wireline data communication, low-jitter clock generation, energy-efficient computing devices for sensor networks, and biosensor applications.

In this talk I will give an overview of empirical software engineering. Then I will discuss two specific research projects I have been working on. The first project is the development of a qualitative methodology for conducting empirical software engineering research. The second project deals with software engineering for high performance computers.

Empirical software engineering is useful or building support for hypotheses. But, the context changes from one study to the next, so it is not always clear how to compare their results. Various results from isolate studies are not as useful as the deeper conclusions that can be drawn when those results are analyzed together. The methodology allows researchers to identify variables, generate hypotheses and build support for those hypotheses. The methodology consists of the following steps, which will be discussed in more detail:

1) Gather expert opinion about potential variables;
2) Develop and refine hypotheses for those variables based on existing data;
3) Run new studies to verify the hypotheses.

After giving an overview of the methodology, I will describe its use for identifying and building support for hypotheses about the relationship of an inspector’s background and experiences to his or her effectiveness during an inspection. The goal of this investigation was to explain the wide variation in the performance of different inspectors.

Finally, I will give an overview of the work I have been doing concerning software engineering for high performance computers. Currently there is very little quantitative understanding of the effort required to develop software for high performance computers or of the tradeoffs between development time and execution time. I will describe a series of planned studies and how we plan to use the results of those studies to generate hypotheses and better understand these issues.

Over the years, probability theory, utility theory and decision theory have emerged as principled frameworks for the design of systems that can adapt, learn and robustly strategize in the presence of uncertainty. In particular, partially observable Markov decision processes (POMDPs) combine these theories into one general framework that can naturally model a wide range of real-word sequential decision problems.

Unfortunately, the considerable expressivity of POMDPs generally results in solution algorithms that are computationally infeasible. In this talk, I will first introduce POMDPs and discuss some important sources of intractability, including the curse of dimensionality as well as the representation of policies and value functions. I will then present a new belief compression technique that mitigates the curse of dimensionality and a bounded policy iteration algorithm that alleviates the complexity of policy search. Both of these approaches allow the solution of large POMDPs more quickly than existing algorithms. Finally, I will discuss the combination of these methods, allowing approximate solutions for problems with belief spaces of dimensionality several orders of magnitude larger than previously addressed in the literature. I will illustrate the application of POMDPs and the above algorithms with a system designed to help elderly persons with various forms of dementia (e.g., Alzheimer's disease) to carry out simple daily tasks.

The current ``best-effort'' Internet, however, does not guarantee Quality of Service (QoS) such as minimum bandwidth, packet loss rate, and delay which are critical to many multimedia applications. As such, many significant challenges remain to design and deploy delay sensitive multimedia applications over the Internet effectively. In this talk, I will present the path diversity (PD) framework for concurrent media streaming to a receiver using multiple routes. Without requiring QoS, the PD framework improves the quality of the streamed media via multiple routes created using either multiple senders or relay nodes, in order to increase available bandwidth, reduce packet loss and delay. The PD framework combats packet loss, delay, and insufficient bandwidth for pre-recorded streaming media by sending packets simultaneously from multiple senders to a single receiver. For interactive and live streaming applications, the PD framework allows a single sender to send packets simultaneously on both default and redundant paths to the receiver. Within the PD framework, I will present a transport protocol to synchronize the simultaneous media streaming to receiver via multiple routes. In particular, the protocol employs the rate allocation and packet partition algorithms. The rate algorithm determines the sending rate on each route in order to minimize the packet loss, while the packet partition algorithm ensures each packet is sent by one and only one sender and at the same time, minimizes the startup delay. I will show theoretically and experimentally that using Forward Error Correction (FEC) in streaming the media simultaneously over multiple mostly independent routes at appropriate sending rates is more effective than using FEC with the traditional uni-path streaming. I will also show the performance gain of coupling Multiple Description Coding (MDC) video with the PD framework over the traditional Single Description Coding (SDC) video using the traditional uni-path approach.

Biography

Thinh Nguyen received his B.S. from University of Washington in 1995, his M.S. and Ph.D. from U.C. Berkeley in 2000 and 2003, respectively. His doctoral work focused on multimedia streaming over the Internet, using the path diversity framework in conjunction with network protocols, and source and channel coding techniques. His research interests include multimedia networking, computer networks, signal processing, machine learning, data analysis and data mining. Prior to graduate school, he worked on computer architecture, advanced graphics and visualization algorithms at Intel Microcomputer Research Lab and at Microsoft Corp. He has won a number of awards including the “Best Paper” award at IEEE Packet Video Workshop for his work on distributed video streaming with forward error correction. More information about him can be found at:
http://www-video.eecs.berkeley.edu/~thinhq

In this talk I'll describe novel learning algorithms for temporal, relational data and their application to trainable video interpretation. I extend an existing visual-event recognition system, Leonard (Siskind 2001), with two new learning components:

1) A new relational sequential inference method that learns mappings from relational observation sequences to relational state sequences. Using this method, we infer force-dynamic world models from raw video, with results comparing favorably to pre-existing hand-coded model reconstructors.

2) A supervised learning method for logical event definitions written in terms of the force-dynamic models constructed in (1) above. I give a specific-to-general learning algorithm that learns definitions that empirically outperform published definitions written by a human expert.

Biography

Alan Fern is a Ph.D candidate in Computer Engineering at Purdue University. He received an M.S degree from Purdue in 2000 and a B.S from the University of Maine in 1997. While at Purdue, Alan received an NSF Graduate Research Fellowship. His primary research interests are in machine learning, data mining, and automated planning/control. He is particularly interested in developing algorithms that leverage rich knowledge representations for learning and discovery in highly structured data.

This talk presents techniques for reducing the power dissipation and chip area of analog filters. The emphasis is on filters used for channel selection in wireless receivers. We first discuss a scheme that uses multiple filtering paths, each optimized for portion of the required total dynamic range. Since the individual filtering paths have small dynamic range they require small power dissipation and chip area. The filtering paths operate all the time allowing the overall system to provide undisturbed output during range-switching, contrary to conventional AGC-filter schemes that give disturbances every time the gain changes. The second part of the talk presents an optimization algorithm for maximizing the dynamic range of an arbitrary filter under a fixed power dissipation constraint. The algorithm uses a simplified Volterra series representation to describe distortion. Closed form expressions are derived for the optimal allocation of power dissipation among the elements of the filter. We next present experimental results from a chip that has been implemented using the proposed techniques. The talk will conclude with a brief overview of the activities of the Communications Circuits Lab of Intel Corporation, Hillsboro, OR. These activities include fully integrated CMOS transceivers, VCOs, synthesizers, very high frequency circuits, etc.

Biography

Yorgos Palaskas received the Diploma in electrical and computer engineering from the National Technical University of Athens, Greece, in 1996, and the M.S. and Ph.D. degrees, both in electrical engineering, from Columbia University, New York, in 1999 and 2002, respectively. His Ph.D. work was in the area of syllabic companding filters for wireless applications. During the summer of 1999 he was with Texas Instruments, NJ, where he worked on disk drive electronics. During the summers of 2000 and 2001 he worked at Agere Systems, New Jersey (formerly Bell Labs), doing research on integrated IF filters. Since January 2003 he has been a Sr. Design Engineer at the Communications and Interconnect Technology Lab of Intel Corporation, Hillsboro, OR, where he is working on wireless transceiver architectures. He holds one US patent and has several others pending.

Attacks against Internet routing are increasing in number and severity. Contributing greatly to these attacks is the absence of origin authentication: there is no way to validate if an entity using an address has the right to do so. This vulnerability is not only a conduit for malicious behavior, but indirectly allows seemingly inconsequential misconfigurations to disrupt large portions of the Internet. This talk discusses the semantics, design, and costs of origin authentication in interdomain routing. A formalization of address usage and delegation is presented and broad classes of cryptographic proof systems appropriate for origin authentication are considered.

The costs of origin authentication are largely determined by the form and stability of the served address space. However, prior to this work, little was known about the relevant characteristics of address use on the Internet. Developed from collected interdomain routing data and presented in this talk, our approximate delegation hierarchy shows that current IP address delegation is dense and relatively static. One notable result shows that as few as 16 entities are the sou