Domain-independent planning is a hard combinatorial problem. Taking into account plan quality makes the task even more difficult. This article introduces Planning by Rewriting (PbR), a new paradigm for efficient high-quality domain-independent planning. PbR exploits declarative plan-rewriting rules and efficient local search techniques to transform an easy-to-generate, but possibly suboptimal, initial plan into a high-quality plan. In addition to addressing the issues of planning efficiency and plan quality, this framework offers a new anytime planning algorithm. We have implemented this planner and applied it to several existing domains. The experimental results show that the PbR approach provides significant savings in planning effort while generating high-quality plans.

This article presents a learning agent shell and methodology for building knowledge bases and agents and their innovative application to the development of a critiquing agent for military courses of action, a challenge problem set by the Defense Advanced Research Projects Agency's High-Performance Knowledge Bases Program. The learning agent shell includes a general problem-solving engine and a general learning engine for a generic knowledge base structured into two main components: (1) an ontology that defines the concepts from an application domain and (2) a set of task-reduction rules expressed with these concepts. The development of the critiquing agent was done by importing ontological knowledge from cyc and teaching the agent how an expert performs the critiquing task. The learning agent shell, the methodology, and the developed critiquer were evaluated in several intensive studies, demonstrating good results.

The Council encouraged Science and Engineering Fair, to be sometimes after an appropriate the Conference Committee to gather held May 8-10 in San Jose. Carol asked waiting period agreeable to our copublisher, extensive feedback after the 2002 conference for a volunteer to replace Mel Montemerlo The MIT Press. The Council voted to gauge how well this new as the coordinator of the judging in favor of reaffirming this policy format was received.

First, we introduce the concept of a learning agent shell as a tool to be used directly by a subjectmatter of theories, methods, and tools that expert (SME) to develop an agent. In his invited talk at the 1993 National strategies. In addition, it supported the (MIT), Stanford University, and Conference on Artificial Intelligence, development of methods for rapidly Northwestern University, developed two Edward Feigenbaum compared the technology extracting knowledge from natural language end-to-end integrated systems that were of a knowledge-based computer texts and the World Wide Web evaluated by Information Extraction system with a tiger in a cage. Rarely does and for knowledge acquisition from subject and Transport Inc. (IET), the challenge a technology arise that offers such a matter experts (SMEs). However, emphasis of the HPKB Program was 1999. Both systems demonstrated high this technology is still far from the use of challenge problems, which are performance through knowledge reuse achieving its potential. This tiger is in a complex, innovative military applications and semantic integration and created a cage, and to free it, the AI research community of AI that are intended to focus the significant amount of reusable knowledge.

The Fourth Symposium on Abstraction, Reformulation, and Approximation (SARA) took place at Horseshoe Bay Resort and Conference Club, Lake LBJ, Texas, from 26 to 29 July 2000. The talks at the conference captured a cross section of application domains and included invited talks and two panels.

The AAAI-2000 Workshop Program was held Sunday and Monday, 3031 July 2000 at the Hyatt Regency Austin and the Austin Convention Center in Austin, Texas. The 15 workshops held were (1) Agent-Oriented Information Systems, (2) Artificial Intelligence and Music, (3) Artificial Intelligence and Web Search, (4) Constraints and AI Planning, (5) Integration of AI and OR: Techniques for Combinatorial Optimization, (6) Intelligent Lessons Learned Systems, (7) Knowledge-Based Electronic Markets, (8) Learning from Imbalanced Data Sets, (9) Learning Statistical Models from Rela-tional Data, (10) Leveraging Probability and Uncertainty in Computation, (11) Mobile Robotic Competition and Exhibition, (12) New Research Problems for Machine Learning, (13) Parallel and Distributed Search for Reasoning, (14) Representational Issues for Real-World Planning Systems, and (15) Spatial and Temporal Granularity.

We are interested in solving real-world planning problems and, to that end, argue for the use of domain knowledge in planning. We believe that the field must develop methods capable of using rich knowledge models to make planning tools useful for complex problems. We discuss the suitability of current planning paradigms for solving these problems. In particular, we compare knowledge rich approaches such as hierarchical task network planning to minimal-knowledge methods such as STRIPS-based planners and disjunctive planners. We argue that the former methods have advantages such as scalability, expressiveness, continuous plan modification during execution, and the ability to interact with humans. However, these planners also have limitations, such as requiring complete domain models and failing to model uncertainty, that often make them inadequate for real-world problems. In this article, we define the terms knowledge-based and primitive-action planning and argue for the use of knowledge-based planning as a paradigm for solving real-world problems. We next summarize some of the characteristics of real-world problems that we are interested in addressing. Several current real-world planning applications are described, focusing on the ways in which knowledge is brought to bear on the planning problem. We describe some existing knowledge-based approaches and then discuss additional capabilities, beyond those available in existing systems, that are needed. Finally, we draw an analogy from the current focus of the planning community on disjunctive planners to the experiences of the machine learning community over the past decade.

Unlike and Exhibition, held 30 July to 3 other contests over the years, there were no August 2000 in Austin, Texas. This year's event artificial walls or constraints in this brought six contest teams and nine exhibition event--the robots had to interact with regular teams from the United States and Canada. Robots were judged on to compete and demonstrate state-ofthe-art the quality of their interactions, coverage, research in robotics and AI (figure 1). An article by the winning team, The competition and exhibition is actually which better describes their approach and made up of multiple events: several contests, a robot, can be found in this issue of AI Magazine. Kortenkamp, Nourbakhsh, and Hinkle (1997); In January 2000, a suggestion was made to Arkin (1998); and Meeden et al. (2000).

In hyperspectral imagery one pixel typically consists of a mixture of the reflectance spectra of several materials, where the mixture coefficients correspond to the abundances of the constituting materials. We assume linear combinations of reflectance spectra with some additive normal sensor noise and derive a probabilistic MAP framework for analyzing hyperspectral data. As the material reflectance characteristics are not know a priori, we face the problem of unsupervised linear unmixing.

The project pursued in this paper is to develop from first information-geometric principles a general method for learning the similarity between text documents. Each individual document is modeled as a memoryless information source. Based on a latent class decomposition of the term-document matrix, a lowdimensional (curved) multinomial subfamily is learned. From this model a canonical similarity function - known as the Fisher kernel - is derived. Our approach can be applied for unsupervised and supervised learning problems alike.