Abstracts of the ISI AI Seminar Series in 2001

Randy Pausch - February 23, 2001

Beverly Park Woolf - March 19, 2001

Kanna Rajan - March 30, 2001

Ken Forbus - April 12, 2001

Kristina Lerman - April 13, 2001

Cheng-Ming Chuong - October 26, 2001

Yolanda Gil - November 9, 2001

Michael Freed - November 19, 2001

Richard Korf - November 30, 2001

Lise Getoor - December 21, 2001

Friday, February 23, 2001

The Interdisciplinary Challenge of Building Virtual Worlds

Randy Pausch

http://www.randypausch.com/

Carnegie Mellon University

Creating an interactive experience is a one of the hardest authoring challenges in human history. Success requires talent from computer science, engineering, art, drama, design, architecture, and a host of other disciplines. Science can tell us how and why to provide better perceptual phenomena (e.g. head-tracking), but success also requires "out of the box" thinking, rapid prototyping tools, and a willingness to create the grammar for this new medium.

Since 1995, I have worked with Walt Disney Imagineering on several virtual reality projects for the DisneyQuest "digital theme park," now open in Orlando and Chicago. Meanwhile, Carnegie Mellon has created the Entertainment Technology Center (ETC) [etc.cmu.edu], a joint initiative between Schools of Computer Science and the College of Fine Arts. As part of the ETC efforts, we have developed the Alice 3d authoring tool [www.alice.org] and processes that allow interdisciplinary teams of undergraduates to build and test compelling virtual worlds in a two-week time period. I will discuss the "Building Virtual Worlds" course, and the mechanisms we use to put students together from different fields.

Biography
Randy Pausch is a Professor of Computer Science, Human-Computer Interaction, and Design at Carnegie Mellon, where he is the co-director of CMU's Entertainment Technology Center (ETC). He was a National Science Foundation Presidential Young Investigator and a Lilly Foundation Teaching Fellow. He has consulted with Walt Disney Imagineering on the user interface design and testing of interactive theme park attractions, particularly for the "DisneyQuest" virtual-reality based theme park. He also consults with Google.com and Z.com on user interface issues. Dr. Pausch is the author or co-author of five books and over 50 reviewed journal and conference proceedings articles, and he sits on the Board of Directors of the Jupiter Media Metrix corporation.

Monday, March 19, 2001

Distributed Interactive Learning Environments

Beverly Park Woolf

http://owl.cs.umass.edu/People/bev/

Center for Knowledge Communication &
Center for Computer-Based Instructional Technology

University of Massachusetts

We describe research directed at building 1) responsive teaching systems with enhanced intelligence and 2) distributed agents and systems in educational marketplaces.

Artificial Intelligence is used to diagnose student learning and provide customized responses. We describe a reinforcement learning architecture that learns to model student performance and then derives a teaching policy to meet a desired educational goal. Spoken natural language is integrated with existing tutors. In addition, rich 3D animation and interactive multimedia are used to support student reasoning, design, and visualization. Psychological studies based on eye-tracking experiments are used to inform spatial reasoning tutors.

Distributed systems include web-based agents that dynamically construct models of available resources, organize massive educational networks and apply resources in support of educational goals. We also describe a web-based homework system that uses interactive discovery environments to provide guided inquiry, along with multimedia simulations and intelligent tutor. The electronic homework system has resulted in improved student grades and reduced faculty costs, sometimes adding a full letter grade to student scores. This system is used by more than 5,000 students at over 20 universities and is being marketed to 100,000 students. We will discuss economies of scale that reduce development costs.

Studies of existing systems help quantifying technology's impact on learning and inform us how to best transition laboratory projects to educational practice.

Bio-sketch
Dr. Beverly Park Woolf is an Associate Research Professor and Director of both the Center for Knowledge Communication and the Center for Computer-based Instructional Technology at the University of Massachusetts. She has a Ph.D. in Computer Science, an Ed.D. in Education and more than 15 years experience in research on educational computer science, production of intelligent tutoring systems and development of multimedia systems. Dr. Woolf is a Fellow of the American Association of Artificial Intelligence (AAAI), an editor for IEEE Computer and an assistant editor for Interactive Learning Environments. She has published over 100 articles and delivered keynote addresses, panels and tutorials in more than 15 countries.

Friday, March 30, 2001

The Remote Agent and Beyond

Kanna Rajan

Autonomy & Robotics Area
Computational Sciences Division
NASA Ames Research Center

As NASA enters the new millennium, the scope of space science and the boundaries of exploration have been pushed tremendously. New missions are rapidly being proposed where round-trip light time and operational difficulties prohibit traditional commanding methods. For instance, NASA's proposed Europa (a Jovian moon) submersible mission requires a spacecraft to land and burrow through miles of what are thought to be ice formations to locate potential life sources around underground volcanic vents. Such a proposed mission cannot be executed currently without more on-board autonomy which place the sense-plan-act loop on a spacecraft in a hostile environment with little observability from ground based controllers.

Yet, another paradigm the agency is pushing is on robotic fleets of explorers cooperatively engaged in science either in planet-finding using interferometry for instance, or for "now-casting" weather information for earth. Controlling such large fleets in a systematic and consistent manner within the current climate of fiscal conservativeness, requires more autonomy, on-board and/or on the ground.

Autonomy is therefore a central tenant at NASA and new funding efforts prove that the agency is serious about deploying missions where the control loop is closed on-board.

The talk will describe the first closed loop on-board autonomy endeavor in space, when the Remote Agent a complex Artificial Intelligence based system controlled NASA's Deep Space One spacecraft. The talk will give an overview of the work with emphasis on a new paradigm on Constraint based Planning techniques for generative planning. We will describe the nature of the planning process and why such a paradigm has been successful despite skepticism from the AI planning community. We will also describe how this foundational work is being branched off for other NASA missions including a proposed ground-based sequencing engine for the most complex NASA mission yet, the Mars 03 Exploration Rovers. We will also describe proposed efforts with Milind Tambe (ISI) on a Distributed Remote Agent being proposed for a NASA testbed.

Biography
Kanna Rajan is a Senior Research Scientist and Group Lead for Spacecraft Autonomy at the Autonomy and Robotics Area at NASA Ames Research Center. He is one of the principals of the Remote Agent (RA) team which designed, built, tested and flew the first AI based closed-loop control system on a spacecraft. The RA was the co-winner of NASA's 1999 Software of the Year award, the agency's highest technical award. His interests are in in Planning/Scheduling, modeling and representation for real world planners. He is the team lead for the Ames Mars03 MER Sequencing project. Prior to joining Ames, he was in the doctoral program at the Courant Institute of Math Sciences at New York Univ. and at the Knowledge Systems group at American Airlines, helping build a Maintenance Routing scheduler which continues to be used by the airline 365 days of the year.

Thursday, April 12, 2001

Towards a computational model of sketching

Ken Forbus

http://www.cs.northwestern.edu/~forbus/forbus.htm

Northwestern University

Sketching is a powerful means of communication between people. While many useful multimodal programs have been created, current systems are far from achieving human-like participation in sketching. A computational model of sketching would help characterize these differences and better understand how to overcome them. The work described in this talk is a first step towards such a model. We start with an example of a sketching system, designed to aid military planners, to provide context. We then describe four dimensions of sketching, visual understanding, conceptual understanding, language understanding, and drawing, that can be used to characterize the competence of existing systems and identify open problems. Three research challenges are posed, to serve as milestones towards a computational model of sketching that can explain and replicate human abilities in this area.

Biography
Kenneth D. Forbus is a Professor of Computer Science and Education at Northwestern University. His research interests include qualitative reasoning, analogy and similarity, cognitive simulation, reasoning system design, articulate educational software, and the use of AI in computer gaming. He received his degrees from MIT (Ph.D. in 1984). He is a Fellow of the American Association for Artificial Intelligence and serves on the Governing Board of the Cognitive Science Society and the editorial boards of Artificial Intelligence, Cognitive Science, and the AAAI Press.

Thursday, April 13, 2001

Multi-agent systems as stochastic systems: a mathematical approach

Kristina Lerman

http://www.isi.edu/~lerman/

USC Information Sciences Institute

Complex collective behavior can often arise out of local interactions between many simple entities, as has been observed in many natural systems. In recent years multi-agent and the robotics communities have begun to explore such biologically inspired systems. These systems offer many advantages over traditional multi-agent systems that are based on deliberative agents and centralized control -- they are efficient, robust, and scalable. Another advantage is that the collective behavior in these systems is often amenable to quantitative mathematical analysis, which can be used to predict behavior of even very large systems and to optimize system performance. While the behavior of an individual agent can be very complex and subject to many unpredictable influences, the behavior of the multi-agent system can often be captured by a simple probabilistic model. We explore a family of such mathematical models that describe the macroscopic or collective dynamics of a multi-agent system. We illustrate our approach by applying it to analyze several agent-based systems. Our examples include coalition formation in an electronic marketplace, and foraging and cooperative task completion in a group of robots.

Biographical information:
Kristina Lerman received her Ph.D. in physics in 1995 from University of California at Santa Barbara, where her research focused on the experimental study of complex dynamical systems. She joined the USC Information Sciences Institute in 1998 as a research scientist. Her current research focuses on two main topics: mathematical modeling of multi-agent systems and applications of machine learning to information extraction and text analysis.

Friday, October 26, 2001

Self Organization of Complex Patterns: A Biologic Case Study on How a Feather is Made

Cheng-Ming Chuong

http://www-hsc.usc.edu/~cmchuong

University of Southern California

How are complex patterns produced in biological and non-biological world? How are the information for the process stored and processed? Here we use the feather as a case study to summarize what we know about the rules of morphogenesis. Feathers comprise one of the most complicated and sophisticated epithelial organs as demonstrated by their shape, size, patterned arrangement, pigmentation, etc. Feathers are not made in one step. Many variations can occur at each of these levels leading to highly complex forms and function. Feathers are progressively patterned during development. The steps include 1) formation of tract fields on the integument surface of the bird, 2) Self organization into periodic arrangement of individual feather primordia within the feather tract, 3) setting up anterior - posterior and proximal - distal axes within a feather primordia, 4) branching morphogenesis of the rachis, barbs and barbules within a feather filament, and 5) additional global processes that produce gradients of variations of morphological parameters or pigmentation across a feather vane or across a whole feather tract.

The breaking of symmetry from the original homogeneous feather primordia initiates the beginning of the making of a feather. Continuous building of multiple levels of asymmetry leads to the complex epithelial appendages we see as feathers. My laboratory has been working on identifying cellular and molecular mechanisms behind the process. As we try to decipher the processes behind many complex patterning existing in the physical world and learn to build complex patterns by genetic engineering and bio-technology, we can certainly learn from this biological case study on the rules of how a feather is built and how the many modulatable morphogenetic process can be used to store a lot of information and provide endless variations in complex pattern formation. It is certainly beneficial for biologists and engineers to learn and appreciate the logic and approaches of each discipline on how to generate complex patterns.

INTRODUCTION OF AUTHOR:
Dr. Cheng-Ming Chuong graduated with a M.D. from Nation of Taiwan University in 1978. He then received his Ph.D. in Developmental and Molecular Biology with Nobel Prize Laureate, Dr. Gerry M. Edelman in Rockefeller University in 1983. He moved to the University of Southern California in 1987 to set up the Laboratory of Tissues Development and Engineering (http://www-hsc.usc.edu/~cmchuong). He is now a Professor of Pathology at University of Southern California.

Dr. Chuong's research has been focused on developmental biology, the knowledge that study how a fertilized egg can develop into a well formed baby, not a tumor mass. He has been focusing to understand the molecular basis and rules used in tissue interactions for the formation of organs, or to examine and analyze the "language" used by the embryo to organize cells into organs. One of the major and unique models used in the laboratory is chicken feather morphgenesis. Using it, they study how the simple epithelial stem cells can generate the most complex patterns. They also studied other epithelial organs including liver, mammary gland, hair, teeth, etc. The goal is to understand fundamental principles of development on the basic side, and to use this knowledge on guiding stem cells for tissue engineering purposes on the application side. Taking a multi-discipline approaches, he also has developed collaborations with paleontologists to search for the origin and evolution of feathers in dinosaurs, and with computer scientists to decipher bio-information stored in biological cells. During these researches, Dr. Chuong's lab also spun out biotechnological inventions including Single Cell CDNA library (patented), RNA Polymerase Chain Reaction (patent pending), DRNA inteference, and other novel technologies which have applications in cancer and many other diseases.

Dr. Chuong is the head of the Tissue Development Engineering Laboratory in Pathology, USC. The laboratory has two faculty (Dr. Wideliez and Dr. Jiang), and more than 15 members with different training background including molecular biology, developmental biology, evolution biology, dermatology, and surgery. Dr. Chuong has approximately 100 publications and one book, "Molecular Basis of Epithelial Appendage Morphogenesis". He is an international leader of his field, and has spoken in and organized many international symposia. He also serves as editors and reviewers for many academic journals, and serve as a reviewer for grant agencies including NIH. His laboratory has been supported by the National Institute of Health, National Science Foundation, and California Breast Cancer Research Program.

Friday, November 9, 2001

Semantic Markup Languages: A Gentle Introduction

Yolanda Gil

http://www.isi.edu/~gil/

USC's Information Sciences Institute

The compelling vision of the Semantic Web ( http://www.w3.org/DesignIssues/Semantic.html, http://www.sciam.com/2001/0501issue/0501berners-lee.html) has motivated a significant amount of research on semantic markup languages. From the W3C, from funding agencies in Europe and the US, and from various research communities, a number of proposals for the underlying representations of the Semantic Web have been put forward. It is likely that many of these languages will be adopted, extended, or dropped (except for some key ideas) by the wider community of developers and users of these languages much as we saw happen to network design and protocols and in the early days of the Internet. This talk will review some of these languages in a tutorial style, and will reflect on their contributions towards the realization of the Semantic Web. This work is in collaboration with Varun Ratnakar.

Monday, November 19, 2001

Simulating Human Agents in Apex

Michael Freed

http://olias.arc.nasa.gov/cognition/people/mike.html

NASA Ames Research Center

In design domains ranging from electronic circuits to airplane wings, computer simulation has become a routine and indispensable part of the design process. However, simulation is almost never used when designing for systems with a "human in the loop" due to the difficulty of modeling those systems' human components. The goal of the Apex project is to produce a practical, simulation-based engineering tool for design in human-machine systems. The talk will overview how Apex addresses three central challenges: generation of skilled behavior, prediction of design-relevant human performance characteristics, and the usability of the engineering tool itself.

Biography:
Michael Freed is a senior research scientist at NASA Ames Research Center where he leads the Apex human simulation project. His research includes human simulation, robot autonomy and human-robot-interaction, all drawing on a core interest in developing AI technologies for behaving intelligently in complex, dynamic environments. He has a B.S. from the University of Massachusetts and a Ph.D. from Northwestern University, both in computer science.

Friday, November 30, 2001

Optimal Bin Packing

Richard E. Korf

University of California at Los Angeles
Computer Science Department

In this talk, I will consider the NP-complete problem of bin packing. Given a set of numbers, and a set of bins of capacity C, find the minimum number of bins needed to contain all the numbers, such that the sum of the numbers assigned to any given bin is less than or equal to C. We will briefly consider two heuristic approximation algorithms for this problem, first-fit decreasing and best-fit decreasing. The main focus on the talk, however, will be a new algorithm for finding optimal or minimum-bin solutions. The algorithm relies heavily on the idea that certain packings of a given bin can never be worse than other packings, and only these undominated packings need be considered. Our algorithm outperforms the best existing optimal bin-packing algorithm by three orders of magnitude in running time.

Biographical Sketch for Richard E. Korf:
Richard Korf is a Professor of computer science at the University of California, Los Angeles. He received his B.S. from M.I.T. in 1977, and his M.S. and Ph.D. from Carnegie-Mellon University in 1980 and 1983, respectively, all in computer science. From 1983 to 1985, he served as Herbert M. Singer Assistant Professor of Computer Science at Columbia University. His research is in the areas of problem solving, planning, and heuristic search in artificial intelligence. He is the author of "Learning to Solve Problems by Searching for Macro-Operators" (Pitman, 1985). He serves on the editorial board of the {\it Journal of Applied Intelligence} and {\it Artificial Intelligence}. Dr. Korf is the recipient of a 1985 IBM Faculty Development Award, a 1986 NSF Presidential Young Investigator Award, the first UCLA Computer Science Department Distinguished Teaching Award in 1989, and the UCLA Engineering School First Annual Student's Choice Award for Excellence in Teaching in 1996. He was elected a Fellow of the American Association for Artificial Intelligence in 1994.

Friday, December 21, 2001

Learning Statistical Models from Relational Data

Lise Getoor

University of Maryland

A large portion of real-world data is stored in commercial relational database systems. In contrast, most statistical learning methods work only with "flat" data representations. Thus, to apply these methods, we are forced to convert the data into a flat form, thereby losing much of the relational structure present in the data and potentially introducing statistical skew. These drawbacks severely limit the ability of current methods to mine relational databases.

In this talk I will review recent work on probabilistic models, including Bayesian networks (BNs) and Probabilistic Relational Models (PRMs), and then describe the development of techniques for automatically inducing PRMs directly from structured data stored in a relational or object-oriented database. These algorithms provide the necessary tools to discover patterns in structured data, and provide new techniques for mining relational data. As we go along, I'll present experimental results in several domains, including a biological domain describing tuberculosis epidemiology, a database of scientific paper author and citation information, and Web data. Finally I will present an application of these techniques to the task of selectivity estimation for database query optimization.

Friday, January 18, 2002

TBD

Dan Suciu

http://www.cs.washington.edu/homes/suciu/

University of Washington

Friday, February 1, 2002

TBD

Charles Elkan

http://www-cse.ucsd.edu/users/elkan/

University of San Diego

Friday, February 15, 2002

TBD

William Swartout

http://www.ict.usc.edu/swartout/

USC's Institute for Creative Technologies

Friday, March 1, 2002

TBD

Sattiraju Prabhakar

http://www.isi.edu/~prabhaka

USC's Information Sciences Institute

[AI Seminar | Schedule for Speaker | Upcoming Seminars ]
[ About AI Seminars | Useful Information ]