Explanation-based Generalization Research Articles

A Semiautomatic Approach to Deriving Turbine Generator Diagnostic Knowledge

Condition monitoring of turbine generators, housed at British Energy nuclear power stations throughout the U.K., is implemented to diagnose incipient faults at an early stage, so corrective action can be taken to avoid the associated high costs of an unplanned shutdown. A prototype expert system has been developed that provides decision support to condition monitoring experts who monitor British Energy turbine generators. The expert system automatically interprets data from strategically positioned sensors and transducers on the turbine generator by applying expert knowledge in the form of heuristic rules. This paper reviews the application domain and describes the work undertaken in developing the prototype expert system. The paper also outlines a learning module design that uses an approach based on an analytical symbolic machine learning technique, explanation-based generalization, to semiautomatically derive heuristic rules for turbine generator fault diagnosis. The approach adopted by the learning module is explained in detail and a worked example demonstrates how the learning module can derive a fault heuristic from a single training example. The modular approach to capturing the causal fault and behavioral models is described, and the method in which the module will be integrated with the existing expert system has been outlined. A preliminary evaluation of the learning module design is discussed.

Explanation based scenario generation for reactive system models

Reactive systems control many useful and complex real-world devices. Tool-supported specification modelling helps software engineers design such systems correctly. One such tool is a scenario generator, which constructs an input event sequence for the spec model that reaches a state satisfying given criteria. It can uncover counterexamples to desired safety properties, explain feature interactions in concrete terms to requirements analysts, and even provide online help to end users learning how to use a system. However, while exhaustive search algorithms work in limited domains, the problem is highly intractable for the functionally rich models that correspond naturally to complex systems engineers wish to design. This paper describes a novel heuristic approach to the problem that is applicable to a large class of infinite state reactive systems. The key idea is to piece together scenarios that achieve subgoals into a single scenario achieving the conjunction of the subgoals. The scenarios are mined from a library captured independently during requirements acquisition. Explanation-based generalization then abstracts them so they may be coinstantiated and interleaved. The approach is implemented, and I present the results of applying the tool to tasks arising from a case study of telephony feature interactions.

Explanation-based interpretation of open-textured concepts in logical models of legislation

In this paper we discuss a view of the Machine Learning technique called Explanation-Based Learning (EBL) or Explanation-Based Generalization (EBG) as a process for the interpretation of vague concepts in logic-based models of law. The open-textured nature of legal terms is a well-known open problem in the building of knowledge-based legal systems. EBG is a technique which creates generalizations of given examples on the basis of background domain knowledge. We relate these two topics by considering EBG's domain knowledge as corresponding to statute law rules, and EBG's training example as corresponding to a precedent case. By making the interpretation of vague predicates as guided by precedent cases, we use EBG as an effective process capable of creating a link between predicates appearing as open-textured concepts in law rules, and predicates appearing as ordinary language wording for stating the facts of a case. Standard EBG algorithms do not change the deductive closure of the domain theory. In the legal context, this is only adequate when concepts vaguely defined in some law rules can be reformulated in terms of other concepts more precisely defined in other rules. We call `theory reformulation' the process adopted in this situation of `complete knowledge'. In many cases, however, statutory law leaves some concepts completely undefined. We then propose extensions to the EBG standard that deal with this situation of `incomplete knowledge', and call `theory revision' the extended process. In order to fill in `knowledge gaps' we consider precedent cases supplemented by additional heuristic information. The extensions proposed treat heuristics represented by abstraction hierarchies with constraints and exceptions. In the paper we also precisely characterize the distinction between theory reformulation and theory revision by stating formal definitions and results, in the context of the Logic Programming theory. We offer this proposal as a possible contribution to cross fertilization between machine learning and legal reasoning methods.

An integration of knowledge acquisition techniques and EBL for real-world production planning

The paper presents an approach to the integration of knowledge acquisition (KA) techniques and explanation-based learning (EBL). Knowledge acquisition techniques are used to delineate a problem class hierarchy for different manufacturing tasks in mechanical engineering. This hierarchy is stepwise formalized into a terminological representation language. The terminological descriptions are then combined with cases of specific manufacturing tasks and their solutions (in the form of production plans). Explanation-based learning is applied to the cases and skeletal plans are automatically constructed for the terminal classes of the problem class hierarchy. Such skeletal plans consist of a dependency structure with a sequence of operators, that can be instantiated to specific plans for all other problems of the class. An evaluation of the proposed KA/EBL integration demonstrates its strengths as well as certain limitations of explanation-based generalization.

A unified framework for explanation-based generalization of partially ordered and partially instantiated plans

Most previous work in explanation-based generalization (EBG) of plans dealt with totally ordered plans. These methods cannot be directly applied to generalizing partially ordered partially instantiated plans, a class of plans that have received significant attention in planning. In this paper we present a natural way of extending the explanation-based generalization methods to partially ordered partially instantiated (POPI) plans. Our development is based on modal truth criteria for POPI plans [3]. We develop explanation structures from these truth criteria, and use them as a common basis to derive a variety of generalization algorithms. Specifically we present algorithms for precondition generalization, order generalization, and possible correctness generalization of POPI plans. The systematic derivation of the generalization algorithms from the modal truth criterion obviates the need for carrying out a separate formal proof of correctness of the EBG algorithms. Our development also systematically explicates the tradeoffs among the spectrum of possible generalizations for POPI plans, and provides an empirical demonstration of the relative utility of EBG in partial ordering, as opposed to total ordering, planning frameworks.

Incorporating Explanation-Based Generalization with Analogical Reasoning

Incorporating Explanation-Based Generalization with Analogical Reasoning

DKBLM — Deep knowledge based learning methodology

To solve the Imperfect Theory Problem (ITP) faced by Explanation Based Generalization (EBG), this paper proposes a methodology, Deep Knowledge Based Learning Methodology (DKBLM) by name, and gives an implementation of DKBLM, called Hierarchically Distributed Learning System (HDLS). As an example of HDLS's application, this paper shows a learning system (MLS) in meteorology domain and its running with a simplified example. DKBLM can acquire experiential knowledge with causality in it. It is applicable to those kinds of domains, in which experiments are relatively difficult to carry out, and in which there exist many available knowledge systems at different levels for the same domain (such as weather forecasting).

Higher-order and modal logic as a framework for explanation-based generalization

Certain tasks, such as formal program development and theorem proving, fundamentally rely upon the manipulation of higher-order objects such as functions and predicates. Computing tools intended to assist in performing these tasks are at present inadequate in both the amount of ‘knowledge’ they contain (i.e., the level of support they provide) and in their ability to ‘learn’ (i.e., their capacity to enhance that support over time). The application of a relevant machine learning technique—explanation-based generalization (EBG)—has thus far been limited to first-order problem representations. We extend EBG to generalize higher-order values, thereby enabling its application to higher-order problem encodings. Logic programming provides a uniform framework in which all aspects of explanation-based generalization and learning may be defined and carried out. First-order Horn logics (e.g., Prolog) are not, however, well suited to higher-order applications. Instead, we employ λProlog, a higher-order logic programming language, as our basic framework for realizing higher-order EBG. In order to capture the distinction between domain theory and training instance upon which EBG relies, we extend λProlog with the necessity operator □ of modal logic. We develop a meta-interpreter realizing EBG for the extended language, λ□Prolog, and provide examples of higher-order EBG.

Higher-Order and Modal Logic as a Framework for Explanation-Based Generalization

Certain tasks, such as formal program development and theorem proving, fundamentally rely upon the manipulation of higher-order objects such as functions and predicates. Computing tools intended to assist in performing these tasks are at present inadequate in both the amount of ‘knowledge’ they contain (i.e., the level of support they provide) and in their ability to ‘learn’ (i.e., their capacity to enhance that support over time). The application of a relevant machine learning technique—explanation-based generalization (EBG)—has thus far been limited to first-order problem representations. We extend EBG to generalize higher-order values, thereby enabling its application to higher-order problem encodings. Logic programming provides a uniform framework in which all aspects of explanation-based generalization and learning may be defined and carried out. First-order Horn logics (e.g., Prolog) are not, however, well suited to higher-order applications. Instead, we employ λProlog, a higher-order logic programming language, as our basic framework for realizing higher-order EBG. In order to capture the distinction between domain theory and training instance upon which EBG relies, we extend λProlog with the necessity operator s of modal logic. We develop a meta-interpreter realizing EBG for the extended language, λs Prolog, and provide examples of higher-order EBG.

A review of learning

AbstractLearning is one of the important research fields in artificial intelligence. This paper begins with an outline of the definitions of learning and intelligence, followed by a discussion of the aims of machine learning as an emerging science, and an historical outline of machine learning. The paper then examines the elements and various classifications of learning, and then introduces a new classification of learning based on the levels of representation and learning as knowledge-, symboland device-level learning. Similarity- and explanation-based generalization and conceptual clustering are described as knowledge level learning methods. Learning in classifiers, genetic algorithms and classifier systems are described as symbol level learning, and neural networks are described as device level systems. In accordance with this classification, methods of learning are described in terms of inputs, learning algorithms or devices, and outputs. Then there follows a discussion on the relationships between knowledge representation and learning, and a discussion on the limits of learning in knowledge systems. The paper concludes with a summary of the results drawn from this review.

An application of explanation-based learning to protocol conformance testing

A method that employs a machine learning technique for the semiautomatic generation of protocol-conformance test-sequence requirements is described. Given a protocol knowledge representation and some high-level nonexecutable descriptions of protocol behavior, a learning algorithm based on extended explanation-based generalization produces conformance test-sequence requirements for a protocol implementation under test. The role of learning is to compile relevant parts of protocol knowledge into behaviors, consequently inferring executable protocol behaviors. This inference makes explicit constraints that are implicit in both the protocol knowledge and the behaviors. It is shown that the approach facilitates the derivation of new operational constraints on protocol behavior. The new constraints lead to new types of protocol behavior, thereby yielding potentially valuable new conformance test cases. An application of the method to the Alternating Bit Protocol (ABP) (a canonical example in protocols research literature) is described.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Acquiring knowledge by explaining observed problem solving

Learning by Understanding Explanations (LBUE) can be an important tool for automatic knowledge acquisition in diagnostic domains. It takes fuller advantage of an expert's knowledge than some more automatic approaches and uses model-based reasoning to reduce brittleness. LBUE specifically integrates the learning of efficiency information as in Explanation Based Generalization (EBG) with further learning of the causal model. Expert protocols are coded into a machine readable form and then input to the LBUE process. The result of learning is that the system becomes capable of solving the same problem in the future and is also able to apply any newly learned causal model information to novel but similar cases.

A Study of Explanation-Based Methods for Inductive Learning

This paper formalizes a new learning-from-examples problem: identifying a correct concept definition from positive examples such that the concept is some specialization of a target concept defined by a domain theory. It describes an empirical study that evaluates three methods for solving this problem: explanation-based generalization (EBG), multiple example explanation-based generalization (mEBG), and a new method, induction over explanations (IOE). The study demonstrates that the two existing methods (EBG and mEBG) exhibit two shortcomings: (a) they rarely identify the correct definition, and (b) they are brittle in that their success depends greatly on the choice of encoding of the domain theory rules. The study demonstrates that the new method, IOE, does not exhibit these shortcomings. This method applies the domain theory to construct explanations from multiple training examples as in mEBG, but forms the concept definition by employing a similarity-based generalization policy over the explanations. IOE has the advantage that an explicit domain theory can be exploited to aid the learning process, the dependence on the initial encoding of the domain theory is significantly reduced, and the correct concepts can be learned from few examples. The study evaluates the methods in the context of an implemented system, called Wyl2, which learns a variety of concepts in chess including “skewer” and “knight-fork.”

Comparative analysis

Comparative analysis is the problem of predicting how a system will react to perturbations in its parameters, and why. For example, comparative analysis could be asked to explain why the period of an oscillating spring/block system would increase if the mass of the block were larger. This paper formalizes the problem of comparative analysis and presents a technique, differential quantitative (DQ) analysis, which solves the task, providing explanations suitable for use by design systems, automated diagnosis, intelligent tutoring systems, and explanation-based generalization. DQ analysis uses inference rules to deduce qualitative information about the relative change of system parameters. Multiple perspectives are used to represent relative change values over intervals of time. Differential analysis has been implemented, tested on a dozen examples, and proven sound. Unfortunately, the technique is incomplete; it always terminates, but does not always return an answer.

Theory-based inductive learning: An integration of symbolic and quantitative methods : Spencer Star Department of Computer Science, Carnegie-Mellon University, Pittsburgh, Pennsylvania 15213

The objective of this paper is to propose a method that will generate a causal explanation of observed events in an uncertain world and then make decisions based on that explanation. Feedback can cause the explanation and decisions to be modified. I call the method Theory-Based Inductive Learning (T-BIL). T-BIL integrates deductive learning, based on a technique called Explanation-Based Generalization (EBG) from the field of machine learning, with inductive learning methods from Bayesian decision theory. T-BIL takes as inputs (1) a decision problem involving a sequence of related decisions over time, (2) a training example of a solution to the decision problem in one period, and (3) the domain theory relevant to the decision problem. T-BIL uses these inputs to construct a probabilistic explanation of why the training example is an instance of a solution to one stage of the sequential decision problem. This explanation is then generalized to cover a more general class of instances and is used as the basis for making the next-stage decisions. As the outcomes of each decision are observed, the explanation is revised, which in turn affects the subsequent decisions. A detailed example is presented that uses T-BIL to solve a very general stochastic adaptive control problem for an autonomous mobile robot.

Book review: LISP LORE: A Guide to Programming the LISP Machine, (2nd Edition) by Hank Bromley and Richard Lamson (Kluwer Academic Publishers)

Machine Learning: A Guide to Current Research provides a collection of current research papers mostly collected form participants at the Third International Machine Learning Workshop held in 1985. These 77 separate papers were collected and reformatted at Rutgers to produce a consistent format and a common bibliography. The range of learning topics include (taken from the preface): analogy, conceptual clustering, explanation-based generalization, incremental learning, inductive inference, learning apprentice systems, machine discovery, theoretical models, and applications. All of these topics, except incremental and apprentice systems are included as heading in the index. Other useful index headings include research institution (a few are missing, primarily those where the several authors are from more than one place) and program name. The papers are short; the average length of five pages per paper, based on the number of papers and the length of the book, overstates their length because of blank pages due to formatting considerations. In the short space for each paper it is not possible to explain in detail how anything is done. There is space only to say what is being done by the system. Thus, the bibliographic references become important for a deeper understanding of individual techniques.

Explanation-based generalization: A unifying view

The problem of formulating general concepts from specific training examples has long been a major focus of machine learning research. While most previous research has focused on empirical methods for generalizing from a large number of training examples using no domain-specific knowledge, in the past few years new methods have been developed for applying domain-specific knowledge to for-mulate valid generalizations from single training examples. The characteristic common to these methods is that their ability to generalize from a single example follows from their ability to explain why the training example is a member of the concept being learned. This paper proposes a general, domain-independent mechanism, called EBG, that unifies previous approaches to explanation-based generalization. The EBG method is illustrated in the context of several example problems, and used to contrast several existing systems for explanation-based generalization. The perspective on explanation-based generalization afforded by this general method is also used to identify open research problems in this area.

Explanation-based generalization: a unifying view

The problem of formulating general concepts from specific training examples has long been a major focus of machine learning research. While most previous research has focused on empirical methods for generalizing from a large number of training examples using no domain-specific knowledge, in the past few years new methods have been developed for applying domain-specific knowledge to formulate valid generalizations from single training examples. The characteristic common to these methods is that their ability to generalize from a single example follows from their ability to explain why the training example is a member of the concept being learned. This paper proposes a general, domain-independent mechanism, called EBG, that unifies previous approaches to explanation-based generalization. The EBG method is illustrated in the context of several example problems, and used to contrast several existing systems for explanation-based generalization. The perspective on explanation-based generalization afforded by this general method is also used to identify open research problems in this area.

Story-telling as planning and learning

The generation of extended plots for melodramatic fiction is an interesting task for Artificial Intelligence research, one that requires the application of generalization techniques to carry out fully. UNIVERSE is a story-telling program that uses plan-like units, ‘plot fragments’, to generate plot outlines. By using a rich library of plot fragments and a well-developed set of characters, UNIVERSE can create a wide range of plot outlines. In this paper we illustrate how UNIVERSE's plot fragment library is used to create plot outlines and how it might be automatically extended using explanation-based generalization methods. Our methods are based on analysis of a television melodrama, including comparisons of similar stories.