If you would like to receive e-mail notification of these seminars please request to be added to the seminars mailing list. A dynamic search feature will be available soon.

RIACS Seminar #100

Date: October 29, 2001

Title: "Necessity is the Mother of Invention"

Speaker(s): James Martin

Affiliation(s): Rockwell Science Center

Abstract:
I will present a brief review of some of the technical projects with which I have been involved in the past, including experimental particle physics, robotics, technology access, and artificial intelligence. The common theme has been: "How do we turn scientific and technical capabilities into something that can be used, or sold as a product?". Critical factors in determining the best approach are understanding the domain of use and getting the full participation of one's partners/customers. Keeping their support requires producing usable and useful deliverables on an on-going basis. Flexibility is essential, especially since requirements evolve and mature during a project. My take-away lesson from the last two decades of R&D bears a great similarity to the old aphorism: Need is very much a primary motivator of innovation.

RIACS Seminar #99

Date: October 23, 2001

Title: "Information Mediation: Integrating Information from Multiple Online Information Sources"

Speaker(s): Dr. Naveen Ashish

Affiliation(s): IBM Silicon Valley Laboratory

Abstract:
This talk centers around "Information Mediators", which are software systems that provided integrated query access to multiple distributed information sources such as databases or Web sources. I will begin with an introduction to such systems, and provide an overview of several interesting research issues in building such systems. I will then discuss in detail an approach to optimizing the performance of these systems by locally storing data at the mediator side. Optimizing performance is the problem that I had addressed for my doctoral dissertation. I will conclude with talking about various applications of information mediators, future work in building mediators systems and finally, an overview of commercial ventures in this area.

RIACS Seminar #98

Date: October 23, 2001

Title: "Do Actions Really Speak Louder than Words?: Eye-Movements During Passive Listening Tasks"

Speaker(s): Rachel Sussman

Affiliation(s): Tanenhaus student at University of Rochester and Stanford University

Abstract:
There is an ever-growing body of work linking eye-movements to language processing in action based tasks. When performing a task within a circumscribed visual world, subjects' eye-movements to objects in the scene reflect their moment-by-moment understanding of auditorily presented instructions. Relatively fewer studies have looked at how language drives eye-movements in more passive listening domains. That is, in the absence of any pressing, real-world task to perform, are eye-movements still dependent on and time-locked to the syntactic processing of spoken language? This talk will give a brief overview of work that examines these questions, as well as present results from a new study that examines eye-movements during interrogatives. The study presented subjects with a spoken narrative and an accompanying display of characters and objects mentioned in the story. After the narrative, subjects were required to verbally answer a "comprehension question" about the story. All of the entities relevant to the question were present in the visual display. Patterns of eye-movements were found to vary depending on the type of question asked (yes/no vs. a question involving a WH-word) as well as the preferred transitivity of the verb in the question. Since subjects were not required to interact with the display in any way, this result provides strong support that eye-movements to visual displays are not merely an upshot of motor planning, but are largely driven by the syntax of linguistic input.

RIACS Seminar #97

Date: October 18, 2001

Title: "Computing with Large Random Patterns: Where AI and Neural Nets May Meet"

Speaker(s): Pentti Kanerva

Affiliation(s): Swedish Institute of Computer Science (SICS)

Abstract:
In traditional computing we calculate with numbers (numeric computing) and we manipulate pointers in computer memory (symbolic computing). These methods have been used in Artificial Intelligence (AI), and impressive results have been achieved in specialized areas. However, we have not been able to build systems that learn in ways resembling human learning, which is the basis of human intelligence. Artificial neural nets are more recent and have had some success with humanlike learning. At SICS we have studied the mathematical foundations of these newer methods that compute with large random patterns, rather than with numbers and pointers. Artificial neural nets are simplified models of real neural circuits. Large patterns are used because the brain's circuits are large--they have thousands to millions of neurons--and because many neurons are active at any one time. The patterns are random because no two brains are wired exactly alike and yet can learn to act alike, for example they can learn to agree about the meanings of words and phrases. This means that each brain has its own internal code based on its peculiar connections. Neural activity patterns are modeled by points of a high-dimensional space, or multicomponent (high-d) vectors where each dimension represents an element of a pattern or a neuron of a circuit. Thus the dimensionality of the vectors is in the thousands to millions. Spaces with such high dimensionality have surprising properties. The study of pattern computing then involves the study of the geometric and algebraic properties of high-dimensional spaces and of computing based on them. Roughly speaking, geometric properties refer to distances between points of the space--i.e., to judgments about similarity--and algebraic properties refer to the making of new meaningful patterns from existing ones. Until recently neurocomputing has relied almost entirely on the geometric properties of the space, which have proven sufficient for tasks such as recognizing handwritten and spoken words. We have focused on the algebraic properties, to provide computational operations for higher mental functions such as learning by analogy and human use of language. In technical terms, the algebraic operations are needed for the encoding and decoding of compositional structure and for analogical mapping of concepts and structures. In this talk I will demonstrate these ideas with the simplest of vectors, namely, with long random bit strings, and indicate what other kinds of vectors and operations have been used.

Speaker's Bio:
Pentti Kanerva has been intrigued by the prospect of understanding brains in computer terms since first learning about computers long ago in Finland. His thesis was the basis of the book Sparse Distributed Memory (MIT Press, 1988). In 1985-92 he was a principal investigator of an SDM research group at RIACS and since then at SICS. Recent work has been on representing compositional structure, and making semantic vectors for words and larger texts, with large random patterns.

RIACS Seminar #96

Date: October 11, 2001

Title: "Packaging Prediction: Prediction-Enabled Component Technologies"

Speaker(s): Dr. Kurt C. Wallnau

Affiliation(s): Carnegie Mellon's Software Engineering Institute

Abstract:
The real innovations of software component technology lie in the areas of 'packaging' and 'deployment'. By focusing on packaging and deployment, it becomes possible to consider how component technology can be used to accelerate experimentation, integration, and transition of prediction technologies. By prediction technology I mean those technologies that permit engineers to analyze and predict the emergent attributes of an assembly of components (security, performance, safety, reliability, etc.) from the known and trusted properties of its constituent components. In this presentation I will explain why progress is possible now, how logical and empirical approaches to certification and prediction can be reconciled, the nature of component interface, the role of certification and trust, and other related ideas. I will then describe a proof of feasibility prediction enabled component technology (PECT) that demonstrates some of the key ideas of predictable assembly. I will close the presentation with a discussion of open issues, and activities that we and others are pursuing to establish a community of collaborators in predictable assembly from certifiable components.

Speaker's Bio:
Kurt C. Wallnau is a senior member of the technical staff at Carnegie Mellon's Software Engineering Institute. Mr. Wallnau currently leads an SEI project in Predictable Assembly from Certifiable Components (PACC). Prior to his work on PACC, Mr. Wallnau led the SEI COTS-Based Systems project. His recent book, Building Systems from Commercial Components (Addison-Wesley) describes the exploratory design processes required by use of commercial components today. Mr. Wallnau's interest in PACC lies in the hope that we can substantially improve the engineering rigor of component-based design processes.

RIACS Seminar #95

Date: October 9, 2001

Title: "A Generic Approach to Monitor Program Executions" (joint work with E. Jahier)

Speaker(s): Mirelle Ducasse

Affiliation(s): INSA/IRISA, Rennes, France

Abstract:
Monitoring requires to gather data about executions. The monitoring functionalities currently available are built on top of ad hoc instrumentations. Most of them are implemented at low-level; in any case they require an in-depth knowledge of the system to instrument. The best people to implement these instrumentations are generally the implementors of the compiler. They, however, cannot decide which data to gather. Indeed, hundreds of variants can be useful and only end-users know what they want.

In this article, we propose a primitive which enables users to easily specify what to monitor. It is built on top of the tracer of the Mercury compiler. We illustrate how to use this primitive on two different kinds of monitoring. Firstly, we implement monitors that collect various kinds of statistics; each of them is well-known, the novelty is that users can get exactly the variants they need. Secondly, we define two notions of test coverage for logic programs and show how to measure coverage rates with our primitive. To our knowledge no definition of test coverage exist for logic programming so far. Each example is only a few lines of Mercury. Measurements show that the performance of the primitive on the above examples is acceptable for an execution of several millions of trace events. Our primitive, although simple, lays the foundation for a generic and powerful monitoring environment.

RIACS Seminar #94

Date: October 4, 2001

Title: "The MER Planning/Scheduling Prototype Effort"

Speaker(s): Dr. Kanna Rajan

Affiliation(s): RIACS, NASA Ames Research Center & The MER P&S Prototyping Team

Abstract:
Since February 01, a joint ARC/JPL team has been developing a software prototype that uses advanced Planning and Scheduling techniques to assist in scheduling of high level science goals for the MER mission. The overall objective of this effort is to ensure that scientists are able to rapidly generate a composite picture of a Sol's activities in an optimal manner consistent with inter-activity constraints and resource bounds. The system operates in a "mixed-initiative" mode that allows user preferences and intent to be fully captured in the planning process while alleviating the burden on the human user of routine tasks such as enforcing flight and mission rules.

This effort combines two systems with a long heritage: JPL's APGEN Mission Planning system; and the Automated Planner/Scheduler flown on-board DS1 as part of the Remote Agent Experiment. It leverages the best of both progenitors by allowing the extensive user interaction facilities in APGEN to be wedded to the constraint maintenance features of the automated planner, providing a tool for rapid scenario analysis and high level science plan generation. Users will see the familiar APGEN interface, but will have additional optional features to automate routine aspects of the science planning process.

In this talk and demonstration of the prototype software, we will show the flexibility the system offers scientists in adding powerful search capabilities to incrementally generate plans, which can then be edited, rearranged or merged with other plans. The talk will focus on the background and the process involved in getting to this stage (including the developmental methodology), and the potential impact this has on MER. The software demonstration will highlight the various features that were incorporated in the system after feedback from a number of MER personnel and project scientists.

The team that developed the prototype includes John Bresina/ARC, Will Edgington/ARC, Ari Jonsson/ARC, Adans Ko/JPL, Pierre Maldague/JPL and Paul Morris/ARC.

RIACS Seminar #93

Date: September 6, 2001

Title: "Spatial Data Sharing and Open GIS Standards"

Speaker(s): Phillip C. Dibner

Affiliation(s): Ecosystem Associates

Abstract:
The ability to share spatially-indexed data has long been sought by governmental agencies, industry, and end users. The petabytes of imagery and other information already online represent an immense resource, much of it in the public trust. As software with spatial components becomes ever more mainstream, the potential for seamless sharing to reduce costs and increase operational capabilities grows accordingly.

However, geospatial datasets are often huge, and not easily amenable to wholesale duplication, conversion, or transfer. The solution lies in interoperability via common interfaces that are independent of underlying storage formats, platforms, or management practices. The growing array of standards developed and published by the Open GIS Consortium (OGC), a partnership of more than 200 vendors, agencies, users, and integrators, addresses this need.

This presentation will describe the standards themselves, the variety of testbeds, pilot projects, and operational installations in which OGC-conforming products have been used, and plans, hopes, and ideas for distributed spatial databanks in the near future.

Speaker's Bio:
Phillip C. Dibner is the principal of Ecosystem Associates, a small consultancy with expertise in software development and environmental analysis.

Mr. Dibner's background combines formal training and professional experience as an ecologist with more than two decades of engineering experience in the software industry, including long stints at Sun Microsystems, NeXT Software, and Apple Computer.

He has been involved with the Open GIS Consortium and its predecessors since 1994. He recently founded, and currently chairs, the OGC's Special Interest Group in Natural Resources and the Environment.

Phillip C. Dibner
Ecosystem Associates
(650) 948-3537
(650) 948-7895 Fax

RIACS Seminar #92

Date: September 6, 2001

Title: "What a J2EE-based Workflow Architecture Can Do for Automated Planning"

Speaker(s): Thomas Hester

Affiliation(s): Bluedot Software

Abstract:
There are some striking similarities between workflows and plans. In fact, a workflow application can be considered to be a plan developer and executor. I will formalize the similarities and difference between workflows and plans. Then I will investigate how workflow applications architecture and concepts may be applied to automated planning.

Contemporary workflow systems execute plans by executing components that represent plan scenarios. These components are generated directly from UML activity diagrams. The components are implemented as Enterprise Java Beans and they are implemented in an application-independent form. The components are then translated into multiple application-dependent forms by XSLT. This approach allows substantial plan reuse and it also supports mixed human and machine execution of plans.

I will present commercial instances of workflow, including one I implemented, and describe specific applications of a corresponding planning system to some NASA problems.

Speaker's Bio:
Thomas R. Hester has worked as a software and systems architect for 20 years. He built natural language understanding and expert systems at Martin Marietta where he was the Program Manager for NASA projects in the Artificial Intelligence Department. At FMC's AI Center, he built route and tactical planners and was Program Manager for DoD projects at the Center. At the Advanced Decision Systems Division of Booz-Allen Hamilton, he was Director of the Planning Practice and created an object model for a planning component of the Air Force air traffic control system. Subsequently, he was Senior Director of the Distributed Objects Practice at Oracle, Chief Technology Officer for FoundryOne and VP of Engineering at Bluedot Software. At FoundryOne he built a workflow system architecture using J2EE components built using Rational Rose, EJB's, BEA's WebLogic and Oracle8 and stored in a UML activity diagram format in a WebDAV repository. Dr. Hester has a PhD in Computational Linguistics from the University of Texas at Austin.

RIACS Seminar #91

Date: August 12, 2002

Title: "SSRP Presentations, Group 4"

Speaker(s) & Title(s):

Vandi Verma of Carnegie Mellon University: "Fault Identification on K9"
Kristin Branson of UC San Diego: "The Earth-Observing Satellite Domain Model"
Brent Venable of University of Padova: "Learning and reasoning with Temporal Soft Constraints"

RIACS Seminar #90

Date: August 23, 2001

Title: "Spatial Data Sharing and Open GIS Standards"

Speaker(s): Dr. Joydeep Ghosh

Affiliation(s): SAIC

Abstract:
Segmentation or clustering of large data sets with thousands of dimensions is very challenging because of sparsity problems arising from the ``curse of dimensionality''. Recent graphical approaches provide novel and efficient solutions by working in similarity space rather than in the original vector space in which the data points were specified. I shall describe one such approach based on constrained, weighted graph-partitioning, that has yielded remarkable results on real-life market basket analysis, clustering web documents and grouping user sessions based on web logs. Issues related to the appropriate choice of similarity space, and how to evaluate and intuitively visualize the clusters will be addressed in the process.

Speaker's Bio:
Joydeep Ghosh is a Full Professor at The university of Texas, Austin, and holder of the Archie Straiton Endowed Fellowship. He directs the Laboratory for Artificial Neural Systems (LANS), where his research group is studying the theory and applications of adaptive pattern recognition, data mining including web mining, and multi-learner systems. Dr. Ghosh has published more than 200 refereed papers and edited 8 books. He has received six best paper awards, including the 1992 Darlington Prize for the Best Journal Paper from IEEE Circuits and Systems Society, and the Best Applications Paper at ANNIE'97. He was a plenary speaker for ANNIE'97 and letters editor for IEEE Transactions on Neural Networks (1998-2000). He is currently an associate editor of Pattern Recognition, Neural Computing Sur veys and Int'l Jl. of Smart Engineering Design.

Dr. Ghosh served as the general chairman for the SPIE/SPSE Conference on Image Processing Architectures, Santa Clara, Feb. 1990 as Conference Co-Chair of Artificial Neural Networks in Engineering (ANNIE)'93 -ANNIE'96, ANNIE '98-2001, and in the program or organizing committees of several conferences on neural networks and parallel processing. More recently, he co-organized workshops on Web Mining (with SIAM Int'l Conf. on Data Mining, 2001) and on Parallel and Distributed Data Mining (with KDD-2000).

RIACS Seminar #89

Date: August 23, 2001

Title: "Data Mining"

Speaker(s): Dr. Raj Kumar

Affiliation(s): The University of Texas, Austin

Abstract:
Data Mining is inductive data analysis. When data is too large and complex to be examined by humans, data mining can produce summaries in the form of a ratio or a formula that reveal patterns. Such patterns enable analysts to come to a better understanding of the data. Data mining can be done on a variety of data types, including numerical, text and spatial data. Data mining technologies provide a spectrum of analytical tools such as classification, segementation and association. However, data mining is still an art: Which analyses to perform in what order is a crtical question. The talk introduces several different data mining algorithms and demonstrates how data mining can be used to analyze datasets. As an application of data mining, the talk will discuss about web data mining and the impact of results to a business. At the end, the pitfalls of data mining will be discussed.

RIACS Seminar #88

Date: August 16, 2001

Title: "SSRP Presentations, Group 3"

Speaker(s) & Title(s):

Jonathan Moody of Carnegie Mellon University: "Making HCC Probabilistic (safely)"
Ellen Campana of University of Rochester: "Gaze and Language: Using Eye Movements to Improve Spoken Dialogue System Performance"
Scott Johnson of University of Wyoming: "Lightweight Temporal Logic Runtime Monitoring"

RIACS Seminar #88

Date: August 9, 2001

Title: "SSRP Presentations, Group 2"

Speaker(s) & Title(s):

Michael Whalen of University of Minnesota: "Software Certification"
Adrian Agogino of University of Texas: "Controlling Collections of Learning Agents"
Kate Mullen of Bard College: "Practical Data Analysis with AutoBayes"

RIACS Seminar #86

Date: August 9, 2001

Title: "Human-Computer Spoken Dialog and Intelligent Tutoring Systems"

Speaker(s): Dr. Gregory Aist

Affiliation(s): Carnegie Mellon University

Abstract:
Human-computer spoken dialog presents the promise of richer interaction with computational systems, as well as access in situations where (for whatever reason) conventional input devices such as mice and keyboards are impractical. Intelligent tutoring systems allow learning to be supported in a computational environment, and to take place where conventional instruction is unavailable. Our long-term goal is to apply human-computer spoken dialog and intelligent tutoring systems to important problems facing society, and furthermore to advance the state of the art in each area. To date we have focused on helping children learn to read. In this talk we will discuss several challenges we faced and how we addressed them. The skills and methods we employed while working on software for children's literacy (as well as in other endeavors) are, however, broadly relevant to ongoing NASA efforts. Furthermore, NASA domains offer important and challenging problems: problems that are worth addressing in their own right, as well as advancing the state of the art in the underlying technologies.

RIACS Seminar #85

Date: August 8, 2001

Title: "Stochastic Anytime Search Algorithms for Nonlinear Optimization and Planning"

Speaker(s): Dr. Benjamin W. Wah

Affiliation(s): Department of Electrical and Computer Engineering University of Illinois, Urbana

Abstract:
In this talk, we present a theory and its associated anytime derivative-free search algorithms for solving problems in autonomous control and planning. Planning and scheduling problems can be defined either by a batch formulation in which a planner plans for one horizon based on its current state and goals, or by a continuous formulation in which a planner uses a current goal set, a plan, a current state, and a model of the expected future state to plan for a much shorter planning horizon. In both cases, an instance of a planning problem can be formulated as a constrained mixed-integer nonlinear programming problem (MINLP) whose objective and constraint functions may not be in closed form or differentiable. We first show the transformation of such a constrained MINLP into an unconstrained penalty function and its solution by unconstrained search algorithms. We focus on characterizing constrained local minima (CLM) in the original MINLP by necessary and sufficient conditions on points in the unconstrained penalty function. Given such one-to-one correspondence, the search for points satisfying the necessary and sufficient conditions in the unconstrained penalty function amounts to locating CLM in the original MINLP. Next, we present efficient heuristics and unconstrained stochastic search algorithms in order to look for points satisfying the conditions. Based on these algorithms, we evaluate stochastic anytime search algorithms that generate solutions of improved quality when given more time and computational resources, and that require the minimum average completion time (up to a constant factor) in order to find solutions of prescribed quality. Although such algorithms do not overcome the NP-hardness of large optimization problems, they lead to solutions of improved quality and constraint satisfaction as more time is spent in solving a problem.

RIACS Seminar #84

Date: August 3, 2001

Title: "Solving Fuzzy Constraint Satisfaction Problems"

Speaker(s): Dr. Hans W. Guesgen

Affiliation(s): University of Auckland, New Zealand

Abstract:
Work in the field of artificial over the past twenty years has shown that many problems can be represented as constraint satisfaction problems and efficiently solved by constraint satisfaction algorithms. However, constraint satisfaction in its pure form isn't always suitable for real world problems, as they often tend to be inconsistent, which means the corresponding constraint satisfaction problems don't have solutions. A way to handle inconsistent constraint satisfaction problems is to make them fuzzy. The idea is to associate fuzzy values with the elements of the constraints, and to combine these fuzzy values in a reasonable way, i.e., a way that directly corresponds to the way how crisp constraint problems are handled. The downside of this approach is that solving the problem becomes more complex. Instead of computing any assignment of values to the variables that satisfies the constraints, we are now confronted with the problem of finding an optimal or at least almost optimal assignment. The purpose of this talk is to discuss various techniques for coping with this task.

Hans Werner Guesgen
Computer Science Department, University of Auckland, New Zealand
hans@cs.auckland.ac.nz
http://www.cs.auckland.ac.nz/~hans
phone +64 (9) 373 7599 ext. 8815, fax +64 (9) 373 7453

RIACS Seminar #83

Date: August 2, 2001

Title: "SSRP Presentations, Group 1"

Speaker(s) & Title(s):

Judah Ben DePaula of University of Texas: "Knowledge Base for Data Tracking"
Alex Groce of Carnegie Mellon University: "Heuristic Search for Model Checking Java Programs"
Matthew Deans of Carnegie Mellon University: "Beating the Kalman Filter"

RIACS Seminar #82

Date: July 26, 2001

Title: "Improving Precision in Information Systems"

Speaker(s): Gio Wiederhold

Affiliation(s): Stanford University

Abstract:
Precision is important when information is to be supplied for commerce and decision-making, However, a major in the web-enabled world is the load and diversity of information. Through the web we can be faced with more alternatives than can be investigated in depth. The value system itself is changing, whereas traditionally information had value, it is now the attention of the purchaser that has value.

New methods and tools are needed to search through the mass of potential information. Traditional information retrieval tools have focused on returning as much possible relevant information, in the process lowering the precision, since much irrelevant material is returned as well. However, for business e-commerce to be effective, one cannot present an excess of data produced. The two types of errors encountered, false positives and false negatives now differ in importance. In most business situations, a modest fraction of missed opportunities (type 1 errors) are acceptable. We will present the tradeoffs and our current direction and tools to enhance precision in information processing to support electronic commerce.

http://www-db.stanford.edu/people/gio.html

RIACS Seminar #81

Date: July 25, 2001

Title: "Computational Studies of the Immune System of The Body"

Speaker(s): Dr. Suhrit Dey

Affiliation(s): Eastern Illinois University

Abstract:
A mathematical model has been developed and solved numerically by perturbed functional iterations. It could predict how cancer strikes the body and how it could grow, especially in the cases of breast cancers. Some reasonable qualitative agreement has been found between mathematical prediction and medical findings.

Suhrit K. Dey, Ph.D
cfskd@eiu.edu

RIACS Seminar #80

Date: July 12, 2001

Title: "Title: 2 + 2 = 4: Shall we prove it?"

Speaker(s): Drs. Claude and Helene Kirchner

Affiliation(s): INRIA, France

Abstract:
Based on joint works with Gilles Dowek and Thérèse Hardin and the Protheo group in Nancy. ``Deduction modulo'' is a way to remove computational arguments from proofs by reasoning modulo a congruence on propositions. Such a technique, issued from automated theorem proving, is of much wider interest because it aims at separating deductions and computations. The first contribution of this talk is to provide a ``sequent calculus modulo'' that gives a clear distinction between the decidable (computation) and the potentially undecidable (deduction).

The congruence on propositions can be handled via rewrite rules and equational axioms. Usually rewriting applies only on terms. We show how to allow rewriting atomic propositions into non atomic ones and to give a complete proof search method, called ``Extended Narrowing and Resolution'' (ENAR), modulo such congruences.

An important application is that this Extended Narrowing and Resolution method subsumes full higher-order resolution when applied to a first-order presentation of higher-order logic. This shows that such a presentation can yield also efficient proof-search methods. If time permits I will show how these concepts could be naturaly implemented in the ELAN framework based on the rewriting calculus.

Keywords: Automated theorem proving, rewriting, resolution, narrowing, higher-order logic, rewriting calculus.

RIACS Seminar #79

Date: July 12, 2001

Title: "Mobile Networking and NASA"

Speaker(s): Marjory Johnson

Affiliation(s): RIACS, NASA Ames Research Center

Abstract:
The objective of the NASA Research and Education Network (NREN) Project is to infuse emerging networking technologies into NASA mission applications. Mobile networking will enable exciting new paradigms for NASA science and engineering, enhancing support for missions that extend into remote areas where it is not economically feasible to create a permanent wired communications infrastructure.

NREN recently hosted a workshop at NASA Ames on "Mobile Terrestrial and Space Networking: Supporting the Scientific Community." The workshop included demonstrations of applications that motivate the need for mobile networking, presentations on satellite communications, wireless networking and sensor networking, and discussions of issues regarding seamless integration. This seminar provides an overview of NREN activities in the area of mobile networking and summarizes the recent workshop demonstrations, presentations, and discussions.

RIACS Seminar #78

Date: June 21, 2001

Title: "Particle Filtering"

Speaker(s): Nando de Frietas

Affiliation(s): U.C. Berkeley

Abstract:
Particle filtering (PF) is a sequential Monte Carlo (SMC) method that goes back to the first publicly available paper in the modern field of Monte Carlo simulation (Metropolis and Ulam, 1949). These algorithms allow us carry out on-line estimation of probability distributions when the data arrives sequentially. Recent methodological and computational advances have led to many success stories in tracking, automatic navigation, financial time series, communications and control. In the context of statistical inference and learning, the performance of PF often deteriorates in high-dimensional state spaces. In the past, we have shown that if a model admits partial analytical tractability, it is possible to combine PF with exact algorithms (Kalman filters, HMM filters, junction tree algorithm) to obtain efficient high dimensional filters. In particular, we exploited a marginalisation technique known as Rao-Blackwellisation. I will provide and introductory review, discuss Rao-Blackwellisation and some new developments.

RIACS Seminar #77

Date: June 21, 2001

Title: "Adventures in Spoken Dialogue Systems"

Speaker(s): John Fry

Affiliation(s): Stanford University

Abstract:
Two different spoken dialogue systems are described, a Japanese Language interface to an office robot, and a tuitorial dialogue system connected to a Navy shipboard crisis simulator. This trains Navy Captains to do refueling in very rough seas. Some of the lessons learned about system architecture and implementations from these two systems will be presented. Future directions for research will be discussed. Finally how spoken dialogue system can benefit from emperical linguistic research on large spoken corpora of transcribed speech will be outlined.

RIACS Seminar #76

Date: May 31, 2001

Title: "Reasoning About Large Systems in a Compositional Way"

Speaker(s): Dimitra Giannakopoulou

Affiliation(s): RIACS, NASA Ames Research Center

Abstract:
Model checking is a highly automated technique fo verifying properties of hardware or software systems. When provided with a formal description of a system's behavior and a required property, the model checker exhaustively explores all the states of the system and returns one of the following results:

* a yes answer, meaning the property is satisfied by the system;
* a counterexample illustrating how the system could violate the property.

More often than not, however, the model checker runs out of memory for any system of realistic size. This problem is known as "state explosion", which refers to the exponential relation between the state space of the system and the components of the state. Various techniques have been proposed to address this problem including abstraction, symbolic state representation, and compositionality.

Compositionality is recognized as the most promising attack to state explosion. It advocates a "divide-and-conquer" approach to verification, and comes in two forms. "Compositional verification" involves reasoning about properties of a system in terms of properties of its components. Compositional minimization exploits the hierarchical structure of a system to incrementally generate and check its behavior. In this talk, we will discuss these two forms of compositionality, and focus on the main benefits and problems that they introduce. We will demonstrate some initial results on applying such techniques to the Remote Agent, a case study that the ASE group has already analyzed in a non-compositional setting.

RIACS Seminar #75

Date: May 24, 2001

Title: "Mulitmodal Spoken Dialogue Systems"

Speaker(s): Joakim Gustafson

Affiliation(s): Telia Research Stockholm, Sweden

Abstract:
Joakim Gustafson, who is visiting Ames from Sweden was formerly at The Royal Institute of Technology (KTH) and is currently employed by Telia Research. This talk will give an overview of several multimodal spoken dialogue systems developed at KTH and Telia Research over the past six years including Waxholm ( http://www.speech.kth.se/waxholm/waxholm2.html ), August (http://www.speech.kth.se/august/ )and Adapt.

RIACS Seminar #74

Date: May 23, 2001

Title: "PV: An On-the-fly LTL-x Model-Checker Combining Selective State Caching and Partial Order Reductions" & "Platform Independent Thread Externalization In Java"

Speaker(s): Gary Lindstrom & Ganesh Gopalakrishnan

Affiliation(s): School of Computing, University of Utah

Abstract:
Partial order reduction requires a proviso to ensure that a process is not continuously enabled in a cycle without being moved in the cycle. We demonstrate that the traditional implementation of the proviso using an in-stack check can lead to un-necessary state explosion. We propose a new partial-order reduction algorithm whose proviso does not use the in-stack check, resulting in a much simpler as well as far more efficient (on our benchmarks) algorithm. Moreover, combining on-the-fly LTL-x model-checking using nested depth-first search and the in-stack proviso is further complicated by the need to communicate extra information between the outer and inner depth-first search phases (as done in SPIN) to avoid soundness problems. All these issues can potentially be further complicated if selective state caching were to be performed. Since our proviso does not depend on the stack, our algorithm can very easily support these combinations. A demo of our model-checker PV based on our algorithm will be presented. [Joint work with Ratan Nalumasu et.al.]

Platform Independent Thread Externalization In Java: The Java programming language is designed to make code mobility practical and convenient. However, it does not standardly support mobility of on-going computations because it lacks the capability to capture, externalize and restore a program's execution state.

We present a novel approach to systematically collapse and externalize the execution state of a Java thread. To ensure both compiler and platform independence, the externalization code is introduced into Java class files via post-processing. Our technique involves introducing a new "shutdown" exception class, and wrapping each program phrase that can encounter such an exception with a handler that externalizes the local stack frame, and then rethrows the exception. Overloaded versions of each restartable method are generated which reverse this process and restore each stack frame with appropriately initialized local variables and temporaries, and resume "forward" execution at the appropriate code point.

We demonstrate the viability of our approach by describing a prototype implementation using the Voyager distributed object system as a framework, without modification or post-processing of the Voyager run time libraries. Under our extension, threads and agents can execute a "moveTo" operation in any control context without loss of control state. Very favorable performance measures are reported, notably in terms of code size expansion and run time overhead introduced by post processing to support thread migration. [Joint work with Wei Tao]

RIACS Seminar #73

Date: May 17, 2001

Title: "AMPHION/NAV: A System for the Deduction-Based Synthesis of State Estimation Software"

Speaker(s): Johann Schumann

Affiliation(s): RIACS, NASA Ames Research Center

Abstract:
All vehicles which are not bound to track or road need a navigation system to determine their exact position and attitude. Modern applications like air traffic or complex space missions have strong requirements with respect to acurracy and reliablity of state estimation software. Despite the fact that the mathematical foundation (Kalman filter) has already been developed in 1960, the development of state estimation software is far from trivial and has caused several mission failures (e.g., unit mismatch in Mars Climate Orbiter).

In this talk, I present AMPHION/NAV, a prototypical system for the automatic synthesis of state estimation software. This tool which has been developed by the ASE group, takes as input a graphical specification describing the sensors, their geometric relationship, and global requirements (e.g., desired coordinate systems). Then, a logic-based synthesis engine fully automatically generates executable C++ code. The underlying proof calculus and the domain theory guarantees consistency between specification, domain theory, and synthesized code; for example, necessary transformations of coordinate systems are introduced transparently and in a correct way.

AMPHION/NAV also generates extensive, hyperlinked documentation for the synthesis task, describing each piece of the software in detail. Thus full traceability between specification and code can be ensured.

RIACS Seminar #72

Date: May 14, 2001

Title: "Modeling Dialogue with Autonomous Systems"

Speaker(s): Oliver Lemon and Stanley Peters

Affiliation(s): Stanford University Center for the Study of Language and Information

Abstract:
Automatic dialogue systems now provide assistance over the phone in managing contacts, trading stocks, and other moderately scriptable interactions. Current research focuses on richer models of dialogue adequate for conversations whose structure cannot be predicted -- for example, conversations between people and autonomous systems such as artificially intelligent tutors or robotic devices operating in dynamic environments.

One of our projects is the development of a multi-modal dialogue system for conversations with an autonomous UAV (a robotic helicopter), as part of the WITAS project (see links below for details). We take this system as our starting point, and describe our work on modeling dialogue interactions as information-state updates, where complex information states represent a wide context of an ongoing conversation. We are particularly concerned with developing models rich enough to support powerful message generation capabilities. The research also aims at implementing mixed initiatives in conversation, and supporting dialogues about the system's tasks and abilities.

Links:
http://www-csli.stanford.edu/semlab/witas/demo1/
http://www.ida.liu.se/ext/witas/

RIACS Seminar #71

Date: May 7, 2001

Title: "Intention Reconsideration"

Speaker(s): Martijn Schut

Affiliation(s): University of Liverpool, UK

Abstract:
Autonomous agents are systems capable of autonomous decision making in real-time environments. Computation is a valuable resource for such decision making, and yet the amount of computation that an autonomous agent may carry out will be limited. It follows that an agent must be equipped with a mechanism that enables it to make the best possible use of the computational resources at its disposal.

In so-called Belief-Desire-Intention (BDI) agents, this decision mechanism, known as the reconsideration policy, manages when an agent reconsiders its currently adopted intentions. The BDI agent model recognizes the primacy of beliefs, desires and intentions in rational action, based on the philosophy of practical reasoning. Intentions direct and constrain an agent's planning process and are as such a useful abstraction for controlling the agent's reasoning. Although intentions persist, they are not permanent, and consequently, an agent has to decide on occasion whether to adopt new intentions or to continue to act upon its current intentions. The main focus of our research has been to investigate how an agent reconsiders its intentions efficiently and effectively.

In this talk, we present two approaches to constructing intention reconsideration policies that are adaptable with respect to the environment in which the agent operates. These approaches are rooted in traditional decision theory; the first is based on discrete deliberation scheduling and the second on partially observable Markov decision processes. We demonstrate empirically that these novel approaches increase and agent's effectiveness in domains with real world characteristics, i.e., dynamic, unpredictable and inaccessible.

RIACS Seminar #70

Date: May 4, 2001

Title: "Deriving and Applying Program Synthesis Calculi"

Speaker(s): David Basin

Affiliation(s): University of Freiburg

Abstract:
Over the last decade I have worked with colleagues on several different projects to develop, implement, and automate the use of calculi for program synthesis and transformation. These projects had different motivations and goals and differed too in the kinds of programs synthesized (e.g., functional programs, logic programs, and even circuit descriptions). However, despite their differences they were all based on three simple ideas. First, calculi can be formally derive in a rich enough logic (e.g., higher-order logic). Second, higher-order resolution is the central mechanism used to synthesize programs during proofs of their correctness. And third, synthesis proofs have a predictable form and can be partially or completely automated. In this talk I explain these ideas and illustrate the general methodology employed.

Speaker's Bio:
David Basin leads the Institute for Software Engineering and Programming Languages. He received his bachelors degree in mathematics from Reed College in 1984, his Ph.D. from Cornell University in 1989, and his Habilitation from the University of Saarbruecken in 1996. His appointments include a postdoctoral research position at the University of Edinburgh (1990 - 1991), and afterwards he led a subgroup, within the programming logics research group, at the Max-Planck-Institut fuer Informatik (1992 - 1997). Since 1997 he is a full professor at the University of Freiburg.

RIACS Seminar #69

Date: May 3, 2001

Title: "Towards More Realistic Natural Language Generation for Spoken Dialogue"

Speaker(s): Amanda Stent

Affiliation(s): University of Rochester

Abstract:
When people talk with computers, they interact very differently from how they interact in conversation with other people. Some of these differences come from people's adapting to limitations in the computer's abilities to interact naturally. If we can make computers interact in a more realistic way, we can reduce the cognitive load on users of dialogue systems.

In this talk, we pose two questions:

* How is language use in human-human spoken dialogue different from other types of language use (e.g. monologue, text, human-computer dialogue)?
* How can we build conversational agents capable of producing more human-like dialogue contributions?

In answer to these questions, we first outline some differences between language use in human-human conversations and that in human-computer conversations. We then present an architecture that facilitates flexible, human-like natural language generation. We discuss its theoretical underpinnings, and describe how we have implemented it in the TRIPS conversational agent at the University of Rochester (Stent 1999; Allen et. al. 2000; Allen, Ferguson and Stent 2001). We briefly outline how we have incorporated multimodality. Finally, we describe the different types of evaluation we are currently conducting and preparing to conduct. 

RIACS Seminar #68

Date: May 3, 2001

Title: "Abstraction and Modular Reasoning for the Verification of Software"

Speaker(s): Corina Pasareanu

Affiliation(s): Kansas State University

Abstract:
Modern software systems, which are often concurrent and distributed, must be extremely reliable and correct. Finite-state verification (FSV) techniques, such as model checking, are emerging as the front-runner in the race to automate high-quality assurance of software. Such techniques exhaustively check a finite-state model of a system for violations of system requirements stated in a complementary formalism, such as assertions or temporal logic formulas.

In the first part of our talk, we will address several of the challenges of building finite-state models of software systems, that are amenable to verification using existing FSV tools. First, the existence of very large or infinite data in software, that comes from (potentially) unbounded data types, makes FSV of software difficult. We consider one method for avoiding this problem: tool support for source-to-source data type abstractions that are used to reduce the data domains of a program to small finite domains. Second, most FSV tools are aimed at reasoning about complete systems, but modern software is, increasingly, built as a collection of independently produced components, which are assembled to achieve a system's requirements. We will describe an automatic technique for building finite-state models of software components that enable modular reasoning, taking into account assumptions about the behavior of the environment in which the components will execute. We will illustrate the application of our approach to FSV of software on a large case study, written in Java.

In the second part of the talk, we will describe several possible extensions of our current research work. In general, the abstractions that we studied, are used for checking universal properties (i.e. properties that hold along every possible execution path). We will show how one technique that we developed for the analysis of counter-examples produced by checks of abstracted programs can be customized to enable verification of existential properties (i.e. properties that hold along some possible execution path). We will also show how to extend the abstraction technique (that handles base types) to more general data structures, using abstractions similar to the ones from shape analysis. An example of heap abstraction is the canonical abstraction of a Java object, induced by the abstract values of non-static fields (the abstraction maps all concrete instances of some Java class that have the same abstract values for all non-static fields, to the same abstract instance).

RIACS Seminar #67

Date: May 3, 2001

Title: "Experiences Acquired in the Design of RoboCup Teams"

Speaker(s): Dr. Stacy Marsella

Affiliation(s): Information Sciences Institute

Abstract:
Increasingly, multi-agent systems are being designed for a variety of complex, dynamic domains. Effective agent interactions in such domains raise some of the most fundamental research challenges for agent-based systems, in teamwork, multi-agent learning and agent modeling. The RoboCup initiative was initiated to foster research in multi-agent systems. It is simultaneously both a research effort and a set of competitions based on the domain of soccer. Since it's inception, it has blossomed into a significant international effort which now includes research on multi-agent modeling in disaster rescue. In this talk, I will discuss the research we have done in the RoboCup simulation league and the general lessons we have extracted from participation in RoboCup competitions. We have fielded two teams and have also used the competitions as a source of data for developing team analysis tools. I will also cover recent efforts in the disaster rescue domain.

RIACS Seminar #66

Date: April 5, 2001

Title: "Challenges of Preserving Computing History: The Computer Museum History Center at Moffett Field "

Speaker(s): Dr. John Toole

Affiliation(s): Computer Museum History Center, NASA Ames Research Center

Abstract:
As a partner in the proposed NASA Research Park, the Computer Museum History Center brings a rich heritage and exciting future to Moffett Field. As an independent non-profit organization, it is dedicated to preserving for posterity the artifacts and stories of the information age. It currently boasts a diverse and internationally respected collection of over 3,000 artifacts, 2,000 films and videotapes, 5,000 photographs, 2,000 linear feet of documentation, and gigabytes of historical software - housed temporarily in two warehouses in front of Hangar 1.
Although everyone recognizes how rapidly the information revolution has been changing our lives over the last 50 years, most are unaware that the history of the Information Age is being lost! When the world looks back 500 years in the future, we will owe it to ourselves and our descendants to tell the story how these 50 years have made such a difference.

The seminar will discuss the background of the Computer Museum History Center; some of our artifacts and why they are important; and discuss some of the interesting challenges of presenting history. How might we present software in exciting ways? How can we address different technical and educational audiences? How can we learn from the past? How can we "capture" history? How can we use the dynamics of research and our industry to help create new solutions? These are other provocative questions will be discussed in the context of being operational in a new world-class facility in 2005.

Speaker's Bio:
As the Executive Director and CEO of The Computer Museum History Center, John C. Toole oversees and drives the overall strategic vision of the museum, and reports directly to the Board of Trustees. In this position, Toole leverages more than 28 years of research and development experience in advanced computing, networking, information technology and microelectronics, culminating in national leadership positions in science and technology management across industry, academia, and government.

Formerly one of two deputy directors at the National Center for Supercomputing Applications (NCSA) at the University of Illinois, Urbana Champaign, Toole oversaw the technical operation and coordination of the National Computational Science Alliance. Prior to the NCSA, Toole was the first fulltime director of the National Coordination Office (NCO) for Computing, Information, and Communications, working for the White House Office of Science and Technology Policy. He also served as executive director for High Performance Computing and Communications for the Defense Advanced Research Projects Agency (DARPA), and as acting director - after several years as program manager and deputy office director - of DARPA's Computing Systems Technology Office (CSTO), which was responsible for advancing computing systems technologies. Toole retired from the U.S. Air Force in 1994 after more than 22 years of service. He holds BS and MSEE degrees from Cornell University.

http://www.computerhistory.org

RIACS Seminar #65

Date: April 5, 2001

Title: "The NASA Astrobiology Institute: An Experiment in Establishing a "Virtual Community"

Speaker(s): Dr. Lisa Faithorn

Affiliation(s): RIACS, NASA Ames Research Center

Abstract:
A key objective for NASA's investment in the NASA Astrobiology Institute (NAI) is the support of innovative collaborative research in the emerging arena of astrobiology, involving scientists from multiple disciplines and different institutions. The catalyzing of collaborative research among distributed groups requires that opportunities be provided for face to face and virtual interaction among scientists who have not historically sought to work together. It also requires active exploration of, and attention to, the behaviors and preferences associated with new forms of productive cooperation. NAI was thus established not only to fund scientific research on existing astrobiological topics but also to catalyze new possibilities for scientific collaboration through the activities it promotes among its Members.

NAI was conceived from the beginning as a "virtual institute" in which a variety of communication and collaboration tools, technologies and processes was to be made available to the geographically dispersed Members in order to facilitate productive engagement. The NAI 'experiment" is now in the third year of its initial five year cycle. There are substantial lessons learned as well as challenges to be addressed regarding what it takes to promote scientific collaboration in astrobiology and develop the technological infrastructure necessary to support an active and successful virtual community of scientific colleagues.

This seminar will provide a brief over view of NAI and focus on some of the lessons learned and current challenges now faced, particularly with regard to communications and collaboration technologies and techniques.

Speaker's Bio:
In her role as NAI Collaborative Research Manager over the last 6 months, Lisa has been involved in the efforts of NAI to further the collaboration functions of the Institute and to critically assess and further develop its technological infrastructure. She brings to this role long-term experience in scientific field research and academic teaching as well as expertise in organizational development and collaborative group facilitation. She also incorporates into her work large-scale organizational and project evaluation efforts. Formerly, she co-founded, and, for fourteen years, directed and taught in the graduate program at the California Institute of Integral Studies in San Francisco. Lisa has conducted field research in Papua New Guinea and northern India, as well as organizational culture studies within the U.S.

RIACS Seminar #64

Date: March 22, 2001

Title: "Digital Libraries = succ(succ(ARPAnet))"

Speaker(s): Dr. Eugene Miya

Affiliation(s): NASA Ames Research Center, Computer History Museum

Abstract:
Starting in 1994, three agencies of the Federal Government (NSF designated as lead, DARPA, and NASA) sought funding for the "next great thing." This was basic, fundamental research. Digital Libraries Initiative, Phase 1 was a four year (FY95-FY98), six university effort with fairly homogeneous funding.

Digital Libraries Initiative, Phase 2 is a five year mix of small and medium- sized projects some with Internationally funded partners. Half a dozen new Federal funding partners have also hopped on board. The Initiative is now too large to survey all projects. This presentation will be an overview of how-to best determine mutual interests. We will cover who the players are, who's not playing, and a few highlights about what is happening (locally: 3 projects at Stanford, 2 Berkeley projects [soon 3], and projects at UCD and UCSB). Some project investigators will likely be visiting Ames in the near future.

While the results of DLI-1 and DLI-2 are regard as a long term benefit, if you use the Google search engine, this was developed under DLI-1. Other things to look for will be new experimental network protocols for search engine query, geographic information systems (GIS), new concepts for old documents, biomedical informatics, turning a video stream into a direct access medium, and more.

RIACS Seminar #63

Date: March 21, 2001

Title: "Approximate Objects and Approximate Theories"

Speaker(s): Dr. John McCarthy

Affiliation(s): Stanford University

Abstract:
We propose to extend the ontology of logical AI to include approximate objects, approximate predicates and approximate theories. Besides the ontology we treat the relations among different approximate theories of the same phenomena.

Approximate predicates can't have complete if-and-only-if definitions and usually don't even have definite extensions. Some approximate concepts can be refined by learning more and some by defining more and some by both, but it isn't possible in general to make them well-defined. Approximate concepts are essential for representing common sense knowledge and doing common sense reasoning. Assertions involving approximate concepts can be represented in mathematical logic.

A sentence involving an approximate concept may have a definite truth value even if the concept is ill-defined. It is definite that Mount Everest was climbed in 1953 even though exactly what rock and ice is included in that mountain is ill-defined. Likewise, it harms a mosquito to be swatted, although we haven't a sharp notion of what it means to harm a mosquito.

The talk treats successively approximate objects, approximate theories, and formalisms for describing how one object or theory approximates another.

RIACS Seminar #62

Date: March 16, 2001

Title: "Acquiring Knowledge from Users: Results and Challenges"

Speaker(s): Dr. Yolanda Gil

Affiliation(s): USC, ISI

Abstract:
Allowing users to update and extend the knowledge in an intelligent system remains a largely unresolved research challenge. I will motivate this need based on practical experiences in several planning task domains. I will review briefly the state of the art in knowledge acquisition research to introduce our approach within the EXPECT project at USC/ISI. EXPECT derives a model of the interdependencies between individual pieces of knowledge and analyzes them to understand how new knowledge added by a user fits in and what additional knowledge needs to be acquired. EXPECT's knowledge bases include ontologies and declarative descriptions of problem solving knowledge. EXPECT's representations include declarative descriptions of problem solving knowledge and are closely integrated with LOOM, a knowledge representation system based on description logic. We have developed several effective techniques that include structuring acquisition dialogues with users through scripts, detecting and resolving errors in a knowledge base through interdependency analysis, and interacting with users in English through structured editors. I will describe briefly several large knowledge bases for planning that we have developed over the last decade and that have enabled us to ground our ideas in practical problems, including a workarounds planning aid that showed the best performance at a DARPA High Performance Knowledge Bases Battlespace Challenge Problem. Finally, I will show results from several user evaluations that we have conducted to test some aspects of EXPECT and that illustrate some of the challenges ahead.

Speaker's Bio:
Yolanda Gil is a Senior Research Scientist and Peoject Leader at USC's Information Sciences Institute and a Research Assistant Professor in the Computer Science Department. She is principal investigator of the EXPECT project, with a research focus on developing of knowledge-based systems with large amounts of background knowledge and on modelling and reusing of problem-solving methods to guide knowledge acquisition. Her current research interests include knowledge acquisition, knowledge-based systems, planning, and the semantic web. She received her Ph.D. in Computer Science from Carnegie Mellon University, and her undergraduate degree from the Polytechnic University of Madrid. She recently received a Best Paper Award at the 2001 conference on Intelligent User Interfaces (IUI), and is co-chair of the new conference on Knowledge Capture (K-CAP).

More information can be found at http://www.isi.edu/~gil.

RIACS Seminar #61

Date: March 8, 2001

Title: "Mobile Information Systems - Mobile IP"

Speaker(s): Charles Perkins

Affiliation(s): Nokia

Abstract:
Mobile IP is under serious consideration in various working groups as a protocol component for a new cellular infrastructure. Groups within the IETF, 3GPP, and 3GPP2 all have related but distinctive perspectives on how to realize the still-nascent potential offered by Mobile IP. In this talk, I will describe some of these recent developments, concentrating on Mobile IPv6 and AAA (Authentication, Authorization, and Accounting). AAA is receiving a lot of attention related to Mobile IP and mobile networking, because service providers need authorization before they can establish a business relationship with mobile computer users that may roam into their area of service. This attention to the profit-making possibilities for mobile networking seems likely to provide a big boost for the deployment of Mobile IP. In this way, AAA will also provide additional impetus for creation of the wireless Internet. All major cellular standardization bodies are making Mobile IP and AAA services an integral part of the new cellular infrastructure. There is also a recognition that IPv6 is crucial for the eventual deployment of billions of IP-addressable wireless devices. I will end this talk by taking a look at the interactions between Mobile IPv6 and AAA, pointing out new areas needing work and making some guesses about the directions that may be taken within the IETF.

Speaker's Bio:
Charles E. Perkins is a Research Fellow at Nokia Research Center, investigating mobile wireless networking and dynamic configuration protocols. He is the editor for several ACM and IEEE journals for areas related to wireless networking. He is serving as document editor for the mobile-IP working group of the Internet Engineering Task Force (IETF), and is author or co-author of standards-track documents in the mobileip, manet, IPv6, and dhc (Dynamic Host Configuration) working groups. Charles has served on the Internet Architecture Board (IAB) of the IETF and on various committees for the National Research Council. He is also associate editor for Mobile Communications and Computing Review, the official publication of ACM SIGMOBILE, and is on the editorial staff for IEEE Internet Computing magazine. Charles has authored and edited books on Mobile IP and Ad Hoc Networking, and has published a number of papers and award winning articles in the areas of mobile networking, ad-hoc networking, route optimization for mobile networking, resource discovery, and automatic configuration for mobile computers. See http://people.nokia.net/~charliep for further details.

RIACS Seminar #60

Date: March 2, 2001

Title: "Efficient Control and Learning in Complex Robotic Systems: Robot Teams and Humanoids on Their Best Behavior"

Speaker(s): Dr. Maja Mataric

Affiliation(s): University of Southern California

Abstract:
Behavior-based control, which exploits the dynamics of collections of concurrent, interacting processes coupled to the external world, is both biologically relevant and effective for problems featuring local information, uncertainty, and non-stationarity. We have developed efficient methods for principled behavior-based control and learning in two problem domains: multi-robot coordination and humanoid imitation. In this talk, we focus on the first domain, and touch briefly on the second.

In the multi-robot domain the key challenges involve reconciling individual and group-level goals and achieving scalable, on-line real-time learning. How to do all of this in a distributed behavior-based way in a timely and consistent fashion? We describe our results in making distributed, behavior-based systems perform in a well-behaved fashion on problems of behavior selection at the individual and group level, communication for dynamic task allocation, and on-line model learning. We describe the use of Pareto-optimality and satisficing to make behavior selection both principled and timely, the robust publish/subscribe messaging paradigm for distributed communication, and augmented Markov models for on-line real-time model building for adaptation. We demonstrate the results of these methods on groups of locally-controlled but globally efficient cooperative mobile robots performing distributed collection, multiple-target-tracking and capture, and coordinated object manipulation.

At the end of the talk we touch on the humanoid control domain, where the key challenges are the high dimensionality of the problem and the choice of representation and modularity that properly integrates the perceptual and motor systems. We describe an imitation learning system that employs direct sensory-motor mappings within the behavior-based framework to address how to understand, segment, and map the observed movement onto the existing motor system. The same structure serves for recognition, classification, prediction, and learning. We demonstrate the results on a 20 degree-of-freedom dynamic humanoid imitating human dance and sports movements from visual data.

Speaker's Bio:
Maja Mataric is an assistant professor in the Computer Science Department and the Neuroscience Program at the University of Southern California and the Director of USC Robotics Research Labs. She joined USC in September 1997, after two and a half years as an assistant professor in the Computer Science Department and the Volen Center for Complex Systems at Brandeis University. She received her PhD in Computer Science and Artificial Intelligence from MIT in 1994, her MS in Computer Science from MIT in 1990, and her BS in Computer Science from the University of Kansas in 1987. She is a recipient of the NSF Career Award, the IEEE Early Career Award, and the MIT TR100 Innovation Award. She has worked at NASA's Jet Propulsion Lab, the Free University of Brussels AI Lab, LEGO Cambridge Research Labs, GTE Research Labs, the Swedish Institute of Computer Science, and ATR Human Information Processing Labs. Her research is in the areas of control and learning in behavior-based multi-robot systems and skill learning by imitation based on sensory-motor primitives.

http://robotics.usc.edu/~maja

RIACS Seminar #59

Date: February 26, 2001

Title: "Dynamic Control of Emergent Behavior in E-Commerce Ecologies"

Speaker(s): Dr. Ron Larsen

Affiliation(s): Maryland Applied Information Technology Initiative

Abstract:
Information Dynamics is an information-centric approach to system design and analysis. While in its early stages of development, it starts from the observation that advances in computing and communications hardware technology are not being matched by advances in design, analysis, implementation, operation, maintenance and support. This creates an imbalance that currently is reflected in softening sales of new equipment and difficulty in designing increasingly complex systems. Information Dynamics seeks an alternative paradigm to the traditional process-centric view of system design.

Information Dynamics explicitly considers the role information plays in a system and, thereby, takes into account what information is needed, when it is required, where it is located, and how it contributes to the operation of the system. Information is treated as a dynamic entity; dynamics (e.g., location, timeliness, value) are explicitly considered. Transformation of information consumes resources (e.g., time, memory, bandwidth). System operation, likewise, consumes resources and feeds back on the dynamics of the information upon which it depends. Information value, or utility, is necessarily associated with a context; utility typically changes with time within a given context, and may be instantaneously different across contexts.

In this seminar, the Information Dynamics framework will be described and illustrated using an agent-based electronic commerce scenario. Ultimately, the intent is to be able to model, understand, and control emergent behavior arising from the interaction of many agents in a networked environment.

Speaker's Bio:
Dr. Larsen is currently the Executive Director of the Maryland Applied Information Technology Initiative (MAITI), a consortium of eight Maryland universities committed to doubling their graduates in information technology by 2004. He is also an affiliate associate professor in the Computer Science Department and a researcher on topics related to digital libraries and networked information systems.

Between 1996 and 1999, Dr. Larsen was the Assistant Director of the Information Technology Office (ITO) at the Defense Advanced Research Projects Agency (DARPA), where he developed and managed the Information Management program and the Translingual Information Detection, Extraction, and Summarization (TIDES) program. He was also responsible for DARPA's involvement in the multi-agency Digital Library Initiatives.

Prior to his tenure at DARPA, Dr. Larsen was the Associate Director of the University of Maryland Libraries, where he led the implementation and deployment of a State-wide library automation system supporting the eleven campuses and two laboratories of the University.

From 1968 to 1985, he was a computer scientist at NASA, developing real time mission support systems, conducting research in computer networking, and developing an agency-wide research program in computer science and automation.

RIACS Seminar #58

Date: February 22, 2001

Title: "Verification and Validation of Autonomous and Adaptive Systems"

Speaker(s): Willem Visser & Charles Pecheur

Affiliation(s): RIACS, Computational Sciences Division, NASA Ames Research Center

Abstract:
This talk will summarize discussions that took place during the RIACS Workshop on the Verification and Validation of Autonomous and Adaptive Systems (Asilomar Conference Center, Pacific Grove, CA, 5-7 Dec 2000).

Discussions include:

V&V of Intelligent Systems: How to verify and validate systems featuring some form of AI-based technique, such as model-based, rule-based or knowledge-based systems.
V&V of Adaptive Systems: How to verify and validate systems featuring adaptive behavior, either in the form of parametric adaptation (e.g. neural nets, reinforcement learning) or control adaptation (e.g. genetic programming).
V&V of Complex Systems: How to verify and validate systems with different interacting parts, either within a given location (e.g. layered control architectures) and among several locations (homogenous or heterogenous multi-agent systems).

See http://ase.arc.nasa.gov/vv2000/asilomar-report.html for more details.

RIACS Seminar #57

Date: February 8, 2001

Title: "Issues in Planning and Scheduling of Earth Observing Satellites"

Speaker(s): Robert Morris

Affiliation(s): RIACS, Computational Sciences Division, NASA Ames Research Center

Abstract:
NASA's growing fleet of Earth-observing satellites employ advanced sensing technology to assist scientists in the fields of meteorology, oceanography, biology, and atmospheric science to better understand how the Earth's systems of air, land, water and life interact with each other. Each satellite's limited resources of power, memory, and sensing instrumentation, as well as ground stations for communicating data and telemetry commands, must be efficiently allocated for the purpose of acquiring, storing, and downlinking high quality images of the earth. With fleets of satellites, there is the added requirement of planning for the coordination of satellites to achieve scientific or operational goals.

This talk will be an overview of the challenges faced by both current and future EOS mission operations planners in order to ensure that mission objectives are achieved. This talk will also describe a current effort by the Planning and Scheduling group at Ames to develop an automated system for planning and scheduling earth observations.

Among the specific topics covered in this talk are the following:

* A characterization of the planning problem for EOS operations, including both long term planning and daily scheduling of scientific and operational activities.
* A description of the current mission operations planning and scheduling process, as exemplified by Landsat 7 misssion operations.
* A survey of current approaches to automated planning and scheduling in the EOS domain.
* A glimpse into the future of EOS missions, which will involve confederations or constellations of spacecraft employing a collection of sensing instruments with different resolution capabilities and, possibly, more powerful on-board processing.

RIACS Seminar #56

Date: January 25, 2001

Title: "Mechanizing Software Development by Refinement"

Speaker(s): Douglas R. Smith

Affiliation(s): Kestrel Institute

Abstract:
This talk presents a mechanizable framework for software development by refinement. The framework is based on a category of specifications. One of the key ideas of Designware is representing knowledge about programming concepts, such as algorithm design and datatype refinement, by means of taxonomies of design theories. The framework is partially implemented in the research systems Specware, Designware, and Planware. Specware provides basic support for composing specifications and refinements, and generating code. Specware is intended to be general-purpose and has found use in industrial settings. Designware extends Specware with taxonomies of software design theories and support for constructing refinements from them. Planware builds on Designware to provide highly automated support for requirements acquisition and synthesis of high-performance scheduling algorithms.

RIACS Seminar #55

Date: January 24, 2001

Title: "A Logical Basis for Component-Based Systems Engineering"

Speaker(s): Manfred Broy

Affiliation(s): Institut für Informatik, Technische Universität München

Abstract:
Items we work with a basic system model and description techniques providing specific views and abstractions of systems such as the interface view, the distribution view, and the state transition view. Each of these views is helpful and has its place in the systems development process. We show how to formalize these views by mathematical and logical means. The development of systems consists in working out these views that lead step by step to an implementation, which in our approach is given by a set of distributed, concurrent, interacting state machines. For large systems, the development is carried through several levels of abstraction. We demonstrate how to formalize the typical steps of the development process and how to express and justify them directly in logic. In particular, we treat three steps of development by refinement: refinement within one level of abstraction, transition from one level of abstraction to the other, implementation by glass box refinement. We introduce refinement relations to capture these three dimensions of the development space. We derive verification rules for the refinement steps. This way, a comprehensive logical basis for the development of systems is provided.

This work was carried out within the Forschungsverbund ForSoft, sponsored by the Bayerische Forschungsstiftung, and the project SysLab sponsored by Siemens-Nixdorf and partially supported by the Deutsche Forschungsgemeinschaft under the Leibniz program.

broy@in.tum.de

RIACS Seminar #54

Date: January 12, 2001

Title: "Robotic Walking, Swimming, and Flying: Combining Sensing and Control in Dynamic Robotic Locomotion"

Speaker(s): James Ostrowski

Affiliation(s): General Robotics, Automation, Sensing, and Perception (GRASP) Laboratory, University of Pennsylvania

Abstract:
In this talk, I will describe a hierarchical framework that we are currently developing for the control and motion planning of a class of dynamic robotic locomotion systems. We study systems whose dynamics possess rotational and translational symmetries, as are found in neutrally buoyant motions of spatial rigid bodies. This work builds upon previous research in nonholonomic systems and geometric mechanics that has led to a single, simplified framework that describes this class of systems, which includes examples such as wheeled mobile robots; undulatory robotic and biological locomotion systems, such as snakes, eels, and paramecia; and the reorientation of satellites and underwater vehicles with attached robotic arms. I will describe our current and previous work on vision-based control of an autonomous blimp-like vehicle, where visual servoing techniques have been developed that combine sensing with the underlying dynamics of the system. This includes more recent extensions for controlling such underactuated systems and dealing with external drift fields such as air currents. I will also highlight current research into vision-based tracking of targets and formation keeping for legged and wheeled robots, as well as closed-loop motion planning using visual feedback for a swimming, eel-like robot.

http://www.cis.upenn.edu/~jpo/home.html

RIACS Seminar #53

Date: January 11, 2001

Title: "Dial Tone for Web Meetings - What does it take?"

Speaker(s):Stewart H. Sonnenfeldt

Affiliation(s): Vice President of Corporate Development, WebEx Communications, Inc.

Abstract:
With the explosive growth of the Internet as the critical medium for the global exchange of knowlege and business activity, today's extended enterprises of customers, suppliers, partners and employees are becoming larger, increasingly complex, and more widely dispersed. The benefit of conducting real-time professional/business communication and collaboration over the web has become a tremendous competitive advantage for enterprises and knowlege professionals. Today's enterprises, however, require a comprehensive network services and applications platform that enables flexible and spontaneous sharing of content and applications along with integrated audio and video conferencing. WebEx has built a unique communications infrastructure based on the T.120 standard. This technology is analogous to telephone switching systems and enables true real-time interactive communication sessions that combine voice, data and video. This platform offers deep communication functionality, solid reliability and massive scalability. The talk will give an overview of the challenges and solutions involved in delivering dial-tone for web meetings.

Stewart H. Sonnenfeldt, Vice President, Corporate Development
WebEx Communications, Inc.
100 Rose Orchard Way, San Jose, California 95134
Direct (408) 435-7283
Fax (408) 435-7004
http://www.webex.com

RIACS Seminar #52

Date: December 7, 2000

Title: "Towards Formal Methods for Rational Agents"

Speaker(s):Marie desJardins

Affiliation(s): SRI International

Abstract:
Using inductive machine learning techniques to construct classification models from large, high-dimensional data sets is a useful way to make predictions in complex domains. However, these models can be difficult for users to understand. In this joint work with Penny Rheingans (University of Maryland, Baltimore County), we have developed visualization methods that help users to understand and analyze the behavior of a learned model, including techniques for high-dimensional data space projection, display of probabilistic predictions, instance mapping, variable/class correlation, and analysis of display space variability. In the talk, I will illustrate these techniques in a census domain, and show how they can be used to give further insights beyond the summaries of model behavior that are provided by commonly used statistical tools.

Speaker's Bio:
Marie desJardins is a senior computer scientist at SRI International. Her ongoing research projects are developing methods for multi-agent planning and negotiation and mixed-initiative planning and knowledge acquisition techniques. Other research interests include probabilistic reasoning, decision theory, and knowledge representation. Dr. desJardins was awarded a Ph.D. in artificial intelligence from the University of California at Berkeley in 1992, where her dissertation presented a model for autonomous machine learning in probabilistic domains. She received her A.B. in engineering / computer science from Harvard University in 1985. She can be reached at
SRI International, 333 Ravenswood Ave., Menlo Park CA 94025; Internet: marie@ai.sri.com.

RIACS Seminar #51

Date: December 4, 2000

Title: "Visualization Techniques for Understanding and Analyzing Learned Models"

Speaker(s): Michael Fisher

Affiliation(s): Department of Computer Science, University of Liverpool, Liverpool UK

Abstract:
Agent-Based systems are beginning to be used in a significant number of areas, and are suggested as providing appropriate solutions for an even wider range of problems. Although there is still considerable debate concerning the detail of what exactly should constitute an agent, there is general agreement that an agent-based approach often provides an appropriate abstraction for modeling and implementing complex systems.

The type of agents that we are concerned with here are typically termed `rational' or `intelligent'. Such agents can be characterized as autonomous components, having their own goals and beliefs and being able reason about their present and future behaviour. Although not widespread in the software industry, such agents are likely to be used increasingly often, especially as more complex domains are considered.

In this talk, I will introduce a logical framework (based on combinations of modal and temporal logics) in which simple rational agents can be described, and will then consider the two questions:

1. how can we reason about agents described using this theory; and
2. how can we implement agents that actually correspond to such logical descriptions?

First, I will show how resolution-based proof approaches can be used in order to mechanize these complex logical theories. Next, I will outline our work on implementing rational agents by directly executing their logical specifications. Finally, I will briefly mention our ongoing/future work on proof and implementation methods for rational agents.

http://www.doc.mmu.ac.uk/STAFF/M.Fisher
Centre for Agent Research and Development:
http://www.card.mmu.ac.uk/
Publications (some online):
http://www.doc.mmu.ac.uk/STAFF/M.Fisher/mdf-pubs/mdf-pubs.html