William S. Havens

Optimal design in collaborative design network

We consider a multiagent system whose task is to aid component-centered design by collaborative designers in a supply chain. In the earlier work, collaborative design networks are proposed as a decision-theoretic framework for such a system. In this work, we analyzes how choice of agent interface affects the computational complexity of collaborative design. Based on the analysis, we proposes a set of algorithms that allow agents to produce an overall optimal design by autonomous local evaluation of local designs. We show that these algorithms reduce the complexity exponentially from that of an exhaustive centralized design.

HIERARCHICAL ARC CONSISTENCY FOR DISJOINT REAL INTERVALS IN CONSTRAINT LOGIC PROGRAMMING

There have been many proposals for adding sound implementations of numeric processing to Prolog. This paper describes an approach to numeric constraint processing which has been implemented in Echidna, a new constraint logic programming (CLP) language. Echidna uses consistency algorithms which can actively process a wider variety of numeric constraints than most other CLP systems, including constraints containing some common nonlinear functions. A unique feature of Echidna is that it implements domains for real‐valued variables with hierarchical data structures and exploits this structure using a hierarchical arc consistency algorithm specialized for numeric constraints. This gives Echidna two advantages over other systems. First, the union of disjoint intervals can be represented directly. Other approaches require trying each disjoint interval in turn during backtrack search. Second, the hierarchical structure facilitates varying the precision of constraint processing. Consequently, it is possible to implement more effective constraint processing control algorithms which avoid unnecessary detailed domain analysis. These advantages distinguish Echidna from other CLP systems for numeric constraint processing.

A Hybrid Schema for Systematic Local Search

We present a new hybrid constraint solving schema which retains some systematicity of constructive search while incorporating the heuristic guidance and lack of commitment to variable assignment of local search. Our method backtracks through a space of complete but possibly inconsistent solutions while supporting the freedom to move arbitrarily under heuristic guidance. The version of the schema described here combines minconflicts local search with conflict-directed backjumping. It is parametrized by a variable ordering relation which controls the order in which the search space is explored. Preliminary experimental results are given comparing two instances of the schema to forward checking with conflict-directed backjumping [17] (FC-CBJ).

An Examination of Probabilistic Value-Ordering Heuristics

Searching for solutions to constraint satisfaction problems (CSPs) is NP-hard in general. Heuristics for variable and value ordering have proven useful in guiding the sestrch towards more fruitful areas of the search space and hence reducing the amount of time spent searching for solutions. Static ordering methods impart an ordering in advance of the search and dynamic ordering methods use information about the state of the search to order values or variables during the search. A well-known static value ordering heuristic guides the search by ordering values based on an estimate of the number of solutions to the problem. This paper compares the performance of several such heuristics and shows that they do not give a significant improvement to a random ordering for hard CSPs. We give a dynamic ordering heuristic which decomposes the CSP into spanning trees and uses Bayesian networks to compute probabilistic approximations based on the current search state. Our empirical results show that this dynamic value ordering heuristic is an improvement for sparsely constrained CSPs and detects insoluble problem instances with fewer backtracks in many cases. However, as the problem density increases, our results show that the dynamic method and static methods do not significantly improve search performance.

A Constraint-Based Reasoning Framework for Behavioural Animation

Behaviour is a reflection of a reasoning process that must deal with constraints imposed by an external environment, internal knowledge and physical structure. This paper proposes a framework for behavioural animation that is based on the next generation of object‐oriented, constraint‐based expert systems technology, and applies a control structure of knowledge agents and knowledge units to determine the behaviour of objects to be animated. Knowledge agents are responsible for planning, plan implementation and information extraction from the environment. The activity of an agent is dependent on the knowledge units ascribed to them by the animator. The interaction between agents and knowledge units is resolved by the reasoning engine, and thus, influences the eventual motion displayed. An example given is NSAIL, a pilot implementation using the model‐based ECHIDNA constraint logic programming shell. With this approach, the motion for a sailing scenario and other behavioural domains can be specified at a high level through the characterization of the knowledge agents.

Extending systematic local search for job shop scheduling problems

Hybrid search methods synthesize desirable aspects of both constructive and local search methods. Constructive methods are systematic and complete, but exhibit poor performance on large problems because bad decisions made early in the search persist for exponentially long times. In contrast, stochastic local search methods are immune to the tyranny of early mistakes. Local search methods replace systematicity with stochastic techniques for diversifying the search. However, the lack of systematicity makes remembering the history of past states problematic. Typically, hybrid methods introduce a stochastic element into a basically constructive search framework. Lynce [6] uses randomized backtracking in a complete boolean satisfiability solver which incorporates clause (nogood) learning to ensure completeness. Jussein & Lhomme [4] perform a constructive search while keeping conflict sets (nogoods) in a Tabu list and backtrack via a stochastic local search in the space of conflict sets. Our method, called Systematic Local Search (SysLS) [3], follows the opposite approach. We incorporate systematicity within an inherently stochastic search method (like [2]). SysLS searches through a space of complete variable assignments and relaxes the requirement for maintaining feasibility. It preserves full freedom to move heuristically in the search space with maximum heuristic information available. While many local search methods easily get trapped in local optima, SysLS records local optima as nogoods in a search memory. Nogoods force the search away from these maximally consistent but unacceptable solutions. Our method is analogous to other diversification mechanisms in local search (eg-Tabu search) but is systematic and inherits the sound resolution rule for nogood learning. In this paper, we extend SysLS for optimization and, in particular, for job shop scheduling problems.

Modelling an academic curriculum plan as a mixed-initiative constraint satisfaction problem

This paper describes a mixed-initiative constraint satisfaction system for planning the academic schedules of university students Our model is distinguished from traditional planning systems by applying mixed-initiative constraint reasoning algorithms which provide flexibility in satisfying individual student preferences and needs The graphical interface emphasizes visualization and direct manipulation capabilities to provide an efficient interactive environment for easy communication between the system and the end user The planning process is split into two phases The first phase builds an initial plan using a systematic search method based on a variant of dynamic backtracking The second phase involves a semi-systematic local search algorithm which supports mixed-initiative user interaction and control of the search process Generated curriculum schedules satisfy both academic program constraints and user constraints and preferences Part of the challenge in curriculum scheduling is handling multiple possible schedules which are equivalent under symmetry We show to overcome these symmetries in the search process Experiments with actual course planning data show that our mixed-initiative systems generates effective curriculum plans efficiently.

Intelligent mediation: an architecture for the real-time allocation of interface resources

Operator interfaces to supervisory-control systems are often highly complex, cumbersome to extract information from, and overwhehningl y verbose in the face of abnormal operating conditions. An oft-cited solution is to replace the conventional operator interface with an intelligent interface; one that mediates control system output to present the operator with intelligently formatted information. While an appealing idea, the question of how to do this remains unanswered: intelligent presentation of control system data is a difficult problem. The task is complex because it requires the realtime allocation of limited interface resources. An expert system architecture and methodology called intelligent mediation – for the real-time allocation of limited interface resources is proposed as a solution to this problem.

NSAIL PLAN: an experience with constraint-based reasoning in planning and animation

A constraint-based reasoning system is used for knowledge representation and reasoning in behavioural animation, specifically in the animation of sailing behaviour. The object-oriented ECHIDNA reasoning and constraint logic programming shell handles the constraints for formulation and execution of plans for intelligent entities. At higher levels of control, the observed motion of an object is a reflection of the reasoning process of an intelligent entity as it reacts to its environment. The environment, internal knowledge and physical structure serve as constraints in developing a plan, which in turn provides additional constraints in the animation of reactive behaviour. An animation approach using constraint-based reasoning is presented focusing on details of the planning process adapted in the approach. The implementation of the NSAIL program reveals further insight into applying this approach towards the development of a high-level intuitive interface for behavioural animation.<<ETX>>

Nicolog: A simple yet powerful cc(FD) language

In this paper, we describe Nicolog, a language with capabilities similar to recently developed constraint logic programming (CLP) languages such as CLP(BNR), clp(FD), and cc(FD). Central to Nicolog are projection constraints (PCs), a sublanguage for compiling and optimizing constraint propagation in numeric and Boolean domains. PCs are an interesting generalization of the indexical constraints introduced in cc(FD) and also found in clp(FD). Nicolog compiles a very general class of built-in constraints into equivalent sets of PCs, allowing an arbitrary mixture of integer (easily extensible to real) and Boolean operations. Nicolog also lets the user program PCs directly, making it possible to implement new sophisticated propagation procedures. We show that PCs are a simple, efficient, and flexible way to implement most of the propagation procedures possible in other FD CLP systems. These include procedures for cardinality, constructive disjunction, implication, and mixed Boolean/numeric constraints. Empirical results with a simple prototype Nicolog implementation based on the WAM architecture show it can solve hard problems with speed comparable to the fastest existing CLP systems.