Search

The goal of the InfoSleuth project at MCC is to exploit and synthesize new technologies into a unified system that retrieves and processes information in an ever-changing network of information sources. InfoSleuth has its roots in the Carnot project at MCC, which specialized in integrating heterogeneous information bases. However, recent emerging technologies such as internetworking and the World Wide Web have significantly expanded the types, availability, and volume of data available to an information management system. Furthermore, in these new environments, there is no formal control over the registration of new information sources, and applications tend to be developed without complete knowledge of the resources that will be available when they are run. Federated database projects such as Carnot that do static data integration do not scale up and do not cope well with this ever-changing environment. On the other hand, recent Web technologies, based on keyword search engines, are scalable but, unlike federated databases, are incapable of accessing information based on concepts. In this experience paper, we describe the architecture, design, and implementation of a working version of InfoSleuth. We show how InfoSleuth integrates new technological developments such as agent technology, domain ontologies, brokerage, and internet computing, in support of mediated interoperation of data and services in a dynamic and open environment. We demonstrate the use of information brokering and domain ontologies as key elements for scalability.

This chapter proposes a multiple-step backtracking mechanism to maintain a tradeoff between replanning and rigid backtracking for exception handling and recovery, thus enabling business process management (BPM) systems to operate robustly even in complex and dynamic environments. The concept of BDI (belief, desire and intention) agent is applied to model and construct the BPM system to inherit its advantages of adaptability and flexibility. Then, the flexible backtracking approach is introduced by utilizing the beneficial features of event-driven and means-end reasoning of BDI agents. Finally, we incorporate open nested transaction model to encapsulate plan execution and backtracking to gain the system level support of concurrency control and automatic recovery. With the ability of reasoning about task characteristics, our approach enables the system to find and commence a suitable plan prior to or in parallel with a compensation process when a failure occurs. This kind of computing allows us to achieve business goals efficiently in the presence of exceptions and failures.

This paper describes an approach to improving the robustness of an agent system by augmenting its failure-handling capabilities. The approach is based on the concept of semantic compensation: "cleaning up" failed or canceled tasks can help agents behave more robustly and predictably at both an individual and system level. Our approach is goal-based, both with respect to defining failure-handling knowledge, and in specifying a failure-handling model that makes use of this knowledge. By abstracting failure-handling above the level of specific actions or task implementations, it is not tied to specific agent architectures or task plans and is more widely applicable. The failure-handling knowledge is employed via a failure-handling support component associated with each agent through a goal-based interface. The use of this component decouples the specification and use of failure-handling information from the specification of the agent's domain problem-solving knowledge, and reduces the failure-handling information that an agent developer needs to provide.

Coalition formation in social networks consisting of a graph of interdependent agents allows many choices of which task to select and with whom to partner in the social network. Agents communicate with agents within n network links in their surrounding network. These agents are considered part of an agent's local neighborhood. Agents maintain a database of skills possessed by agents in their local neighborhood. We introduce agents who use the information gathered from their local neighborhood to modify their network links to discover effective network structures. Egalitarian agents seek equal distribution of connections among agents, resulting in a dense network structure. Inventory agents seek to connect to agents who possess a skill not found in their current local neighborhood.

This paper presents the ARTS (Agent-oriented Robust Transactional System) model, which applies transaction concepts to provide agent developers with high-level support for agent system robustness and reliability. ARTS abstractly considers agents as executors of encapsulated task entities which comply with a set of execution constraints on both normative execution and compensation (repair) semantics. ARTS then defines the task interface in terms of predictable terminating states to support a contract-like interaction among agents. In conjunction with this encapsulation of task semantics, ARTS defines a model for specifying scoped compensation and exception-handling plans for a given task, and for systematically selecting and executing these plans --- triggered by subtask events --- so that the enclosing task semantics are enforced. These capabilities together define a model that reduces design complexity while increasing system robustness, by allowing an agent developer to compose recursively-defined, atomically-handled tasks.

In an agent system, the ability to handle problems and recover from them is important in sustaining stability and providing robustness. We claim that execution logging is essential to support agent system robustness, and that agents should have architectural-level support for logging and recovery methods. We describe an infrastructure-level, default methodology for agent problem-handling, based on logging, and supported by declaratively encoding domain-specific knowledge related to changes in goal status and semantic compensations. Via logging, the approach allows repair of already-completed as well as current goals. We define a language, APLR, to support and constrain incremental specification of problem-handling information, with the agents' problem-handling behaviour increasing in so phistication as more knowledge is added to the system. The approach is implemented by mapping the methodology and domain knowledge to 3APL-like plan rules extended to support logging.

Increasing expense of developing microelectronic manufacturing technology threatens to slow the growth of the electronics industry. This paper describes the progress we have made in developing methodologies and techniques to reduce the cost of designing microelectronic manufacturing flows. Our approach is to partition the task of process flow design into a number of abstraction levels and provide mechanisms to translate between these levels. This approach results in a top-down design methodology where requirements from higher levels of abstraction are successively reduced to lower abstraction levels, while meeting the constraints imposed by the lower levels. The paper enumerates the abstraction levels we have identified so far, and describes the translation mechanisms for a class of process design tasks: modification of an existing flow in response to change in performance requirements. Finally, we briefly describe a design environment that incorporates these ideas.

In agent systems, different (autonomous) agents collaborate to execute complex tasks. Each agent provides a set of useful capabilities, and the agent system combines these capabilities as needed to perform complex tasks, based on the requests input into the system. Agent communication languages (ACLs) allow agents to communicate with each other about how to partition these tasks, and to specify the responsibilities of the individual agents that are invoked. Current ACLs make certain assumptions about the agent system, such as the stability of the agents, the lifetime of the tasks and the intelligence of the agents in the system, etc. These assumptions are not always applicable in information-centric applications, since such agent systems contain unreliable agents, very long running tasks, agents with widely varying levels of sophistication, etc. Furthermore, not all agents may be able to support intelligent planning towork around these issues, and precanned interactions used in more component-based systems do not work well. Thus, it becomes important that proper support for task coordination be available to these agent systems. In this paper we explore issues related to coordinating large, complex, long running tasks in agent systems. We divide these issues into the following categories: tasks, roles, and conversations. We then discuss how these issues impose requirements on ACL, and propose changes to support these requirements.

Conversation policies codify allowable exchanges of speech acts among agents as they execute specific types of tasks. Both the set of agents in a community, and the nature of those agents may change over time; however, these agents must conform to a common set of conversation policies that are robust to change and failure. We describe aspects of the implementation of conversation policies in InfoSleuth, including the integral use of finite-state automata for defining those policies. We identify features of those automata and the underlying performatives that are necessary for their robust and correct execution in an operational community. We describe the construction of new conversation policies from simpler underlying components using two mechanisms, extension and concatenation. In this way, we can ensure that the specification of these new policies is easily sharable, and that certain shared characteristics of multiple conversation policies are enforced consistently.

Publisher Summary This chapter discusses bias in planning and explanation-based learning. Bias is any basis for choosing one generalization over another, other than strict consistency with the observed training instances. One has taken seriously the notion of bias as a means of characterizing variations among planners. It is hypothesized that this would yield three benefits: (1) help organize and understand many of the concepts in planning, (2) reduce the computational requirements of planning, and (3) induce effective biases in learning. The first benefit has not yet led to a complete theory or taxonomy of planning methods, but it has led to the development of several orthogonal bias dimensions that provide fragments of organization over the space of methods. Regarding the second benefit, initial experiments with the six planners suggest monotonie trade-offs between completeness and efficiency as the bias dimensions are traversed from least to most restrictive. On the third benefit, the effects of changes in planning bias on bias in explanation-based learning were investigated with a case study in generalization to N.

During the course of the Microelectronics Manufacturing Science and Technology (MMST) program, a number of sensors have been integrated into various test‐bed plasma etch reactors with the goal of monitoring, diagnosing, and controlling these processes. These sensors include single wavelength and spectral ellipsometers for real‐time in situ etch rate and endpoint determination; a standard monochromator for etch rate and nonuniformity measurements; an eddy current sensor for incoming metal thickness control; a rf monitor for rf current and voltage diagnostics; and a scatterometry‐based critical dimension sensor for linewidth measurements. The full integration of these sensors turned out to be a complex and time‐consuming task including hardware, optical, software, material, and processing issues. Once integrated into the reactor, and the appropriate process modeling completed, these sensors enabled the monitoring, diagnosis, and model‐based control of these processes. Besides maintaining specific process ob...

Soar is an architecture for intelligence that integrates learning into all of its problem-solving behavior. The learning mechanism, chunking, has been studied experimentally in a broad range of tasks and situations. This paper summarizes the research on chucking in Soar, covering the effects of chunking in different tasks, task-independent applications of chunking and our theoretical analyses of effects and limits of chunking. We discuss what and when Soar has been able to learn so far. The results demonstrate that the variety of learning in Soar arises from variety in problem solving, rather than from variety in architectural mechanisms. Keywords: Artificial intelligence, Machine learning, cognitive architecture.