Alan K. Mackworth

Consistency in Networks of Relations

Artificial intelligence tasks which can be formulated as constraint satisfaction problems, with which this paper is for the most part concerned, are usually by solved backtracking the examining the thrashing behavior that nearly always accompanies backtracking, identifying three of its causes and proposing remedies for them we are led to a class of algorithms whoch can profitably be used to eliminate local (node, arc and path) inconsistencies before any attempt is made to construct a complete solution. A more general paradigm for attacking these tasks is the altenation of constraint manipulation and case analysis producing an OR problem graph which may be searched in any of the usual ways. Many authors, particularly Montanari and Waltz, have contributed to the development of these ideas; a secondary aim of this paper is to trace that history. The primary aim is to provide an accessible, unified framework, within which to present the algorithms including a new path consistency algorithm, to discuss their relationships and the may applications, both realized and potential of network consistency algorithms.

A theory of multiscale, curvature-based shape representation for planar curves

A shape representation technique suitable for tasks that call for recognition of a noisy curve of arbitrary shape at an arbitrary scale or orientation is presented. The method rests on the describing a curve at varying levels of detail using features that are invariant with respect to transformations that do not change the shape of the curve. Three different ways of computing the representation are described. They result in three different representations: the curvature scale space image, the renormalized curvature scale space image, and the resampled curvature scale space image. The process of describing a curve at increasing levels of abstraction is referred to as the evolution or arc length evolution of that curve. Several evolution and arc length evolution properties of planar curves are discussed. >

The complexity of some polynomial network consistency algorithms for constraint satisfaction problems

Abstract Constraint satisfaction problems play a central role in artificial intelligence. A class of network consistency algorithms for eliminating local inconsistencies in such problems has previously been described. We analyze the time complexity of several node, arc and path consistency algorithms and prove that arc consistency is achievable in time linear in the number of binary constraints. The Waltz filtering algorithm is a special case of the arc consistency algorithm. In the edge labelling computational vision application the constraint graph is planar and so the time complexity is linear in the number of variables.

Characterizing diagnoses and systems

Abstract Most approaches to model-based diagnosis describe a diagnosis for a system as a set of failing components that explains the symptoms. In order to characterize the typically very large number of diagnoses, usually only the minimal such sets of failing components are represented. This method of characterizing all diagnoses is inadequate in general, in part because not every superset of the faulty components of a diagnosis necessarily provides a diagnosis. In this paper we analyze the concept of diagnosis in depth exploiting the notions of implicate/implicant and prime implicate/implicant. We use these notions to consider two alternative approaches for addressing the inadequacy of the concept of minimal diagnosis. First, we propose a new concept, that of kernel diagnosis, which is free of this problem with minimal diagnosis. This concept is useful to both the consistency and abductive views of diagnosis. Second, we consider restricting the axioms used to describe the system to ensure that the concept of minimal diagnosis is adequate.

A logical framework for depiction and image interpretation

We propose a logical framework for depiction and interpretation that formalizes image domain knowledge, scene domain knowledge and the depiction mapping between the image and scene domains. This framework requires three sets of axioms: image axioms, scene axioms and depiction axioms. An interpretation of an image is defined to be a logical model of these axioms. \n The approach is illustrated by a case study, a reconstruction in first order logic of a simplified map understanding program, Mapsee. The reconstruction starts with a description of the map and a specification of general knowledge of maps, geographic objects and their depiction relationships. For the simple map world we show how the task level specification may be refined to a provably correct implementation by applying model-preserving transformations to the initial logical representation to produce a set of propositional formulas. The implementation may use known constraint satisfaction techniques to find the set of models of these propositional formulas. In addition, we sketch preliminary logical treatments for image queries, contingent scene knowledge, ambiguity in image description, occlusion, complex objects, preferred interpretations and image synthesis. \n This approach provides a formal framework for analyzing and going beyond existing systems such as Mapsee, and for understanding the use of constraint satisfaction techniques. It can be used as a foundation for the specification, design and implementation of vision and graphics systems that are correct with respect to the task and algorithm levels.

Model-Driven Interpretation in Intelligent Vision Systems

With a constructive knowledge-based theory of perception as its foundation, this paper starts with a review and critique of some artificial-intelligence programs that purport to see. It is then argued that these computer programs for scene analysis offer the hope of providing a more adequate account of human competence in interpreting line drawings as polyhedra than do the current psychological theories. This thesis has several aspects. The one emphasized here is that those programs have explored a variety of methods of incorporating a priori knowledge of objects through the use of models. After outlining the range of models used, presenting a set of criteria for evaluating the use of model information, and sketching some psychological theories, the various proposals are contrasted. This discussion leads to two new proposals for exploiting model information that involve elaborations of an existing program, POLY.

Hierarchical arc consistency: exploiting structured domains in constraint satisfaction problems

Constraint satisfaction problems can be solved by network consistency algorithms that eliminate local inconsistencies before constructing global solutions. We describe a new algorithm that is useful when the variable domains can be structured hierarchically into recursive subsets with common properties and common relationships to subsets of the domain values for related variables. The algorithm, HAC, uses a technique known as hierarchical arc consistency. Its performance is analyzed theoretically and the conditions under which it is an improvement are outlined. The use of HAC in a program for understanding sketch maps, Mapsee3, is briefly discussed and experimental results consistent with the theory are reported.

Knowledge structuring and constraint satisfaction: the Mapsee approach

Schema-based representations for visual knowledge are integrated with constraint satisfaction techniques. This integration is discussed in a progression of three sketch map interpretation programs: Mapsee-1, Mapsee-2, and Mapsee-3. The programs are evaluated by the criteria of descriptive and procedural adequacy. The evaluation indicates that a schema-based representation used in combination with a hierarchical arc-consistency algorithm constitutes a modular, efficient, and effective approach to the structured representation of visual knowledge. The schemata used in this representation are embedded in composition and specialization hierarchies. Specialization hierarchies are further expanded into discrimination graphs. >

Automated Spatial-Semantic Modeling with Applications to Place Labeling and Informed Search

This paper presents a spatial-semantic modeling system featuringautomated learning of object-place relations from an online annotateddatabase, and the application of these relations to a variety ofreal-world tasks. The system is able to label novel scenes with placeinformation, as we demonstrate on test scenes drawn from the same sourceas our training set. We have designed our system for future enhancementof a robot platform that performs state-of-the-art object recognitionand creates object maps of realistic environments. In this context, wedemonstrate the use of spatial-semantic information to performclustering and place labeling of object maps obtained from real homes.This place information is fed back into the robot system to inform anobject search planner about likely locations of a query object. As awhole, this system represents a new level in spatial reasoning andsemantic understanding for a physical platform.

Improving gene recognition accuracy by combining predictions from two gene-finding programs

MOTIVATION Despite constant improvements in prediction accuracy, gene-finding programs are still unable to provide automatic gene discovery with desired correctness. The current programs can identify up to 75% of exons correctly and less than 50% of predicted gene structures correspond to actual genes. New approaches to computational gene-finding are clearly needed. RESULTS In this paper we have explored the benefits of combining predictions from already existing gene prediction programs. We have introduced three novel methods for combining predictions from programs Genscan and HMMgene. The methods primarily aim to improve exon level accuracy of gene-finding by identifying more probable exon boundaries and by eliminating false positive exon predictions. This approach results in improved accuracy at both the nucleotide and exon level, especially the latter, where the average improvement on the newly assembled dataset is 7.9% compared to the best result obtained by Genscan and HMMgene. When tested on a long genomic multi-gene sequence, our method that maintains reading frame consistency improved nucleotide level specificity by 21.0% and exon level specificity by 32.5% compared to the best result obtained by either of the two programs individually. AVAILABILITY The scripts implementing our methods are available from http://www.cs.ubc.ca/labs/beta/genefinding/