BRATKO, I., MOZETI C, I., AND LAVRA C, N. 1989. KARDIO:A Study in Deep and Qualitative Knowledge for Expert Systems. MIT Press, Cambridge, MA.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....matcher [Dou96] We illustrate our approach on the discovery of expert rules for recognizing cardiac arrhythmias from electrocardiograms (ECG) CCQ 99] Several attempts to learn knowledge for monitoring cardiac patients have already been made. They use formalisms such as expert rules [BML89] neural networks [Wat95] attributed grammars [Sko90,KAG97] or fuzzy rules [KNB98] The originality of our approach lies in the fact that we use symbolic events and explicit temporal constraints that makes the knowledge more readable and understandable by clinical sta#. In the next section we ....

I. Bratko, I. Mozetic, and N. Lavrac. Kardio: A Study in Deep and Qualitative Knowledge for Expert Systems. MIT Press, 1989.

....(ECGs) and the situations to recognize are cardiac arrhythmias. As temporal relations among events are crucial as well as a specification language which can lead to informative explanations, we have chosen to use inductive logic programming (ILP) This is a major di#erence between Kardio [1] and our own approach. Kardio uses feature based induction, thus, it can only learn predefined propositional structural relations. Target concepts are represented as first order formulas in ILP. This makes the rules more Figure1. A normal ECG (on the left) and a bigeminy ECG (on the right) ....

.... 16, 0, 0] 18, 19, 2, 18, 17] p wave(P0, normal, equal(P0, R0) p wave(P1, normal, R0) qrs(R1, normal, P1) class(mobitz2) 0, 0, 2, 0, 0] 18, 19, 16, 18, 17] p wave(P0, normal, equal(P0, R0) p wave(P1, normal, R0) qrs(R1, abnormal, P1) class(normal) 0, 0, 0, 17, 4] [18, 19, 18, 1, 13] p wave(P0, normal, qrs(R0, normal, P0) p wave(P1, normal, R0) qrs(R1, normal, P1) p wave(P2, normal, R1) qrs(R2, normal, P2) p wave(P3, normal, R2) qrs(R3, normal, P3) p wave(P4, normal, R3) class(pvc) 0, 0, 0, 0, 17] 18, 19, 18, 18, 0] p wave(P0, normal, qrs(R0, ....

[Article contains additional citation context not shown here]

I. Bratko, I. Mozetic, and N. Lavrac. Kardio: A Study in Deep and Qualitative Knowledge for Expert Systems. MIT Press, 1989.

....4 : Identifying function differences Figure 5 : List of candidates system would guide the user through the possible diagnoses. This is not a completely novel strategy and builds on previous work in generating fault dictionaries by simulation. Previous work which stands out in this area includes (Bratko et al. 1989) and (Mauss and Neumann 1995) However, there are two significant differences between this work and those previous efforts to build diagnostic fault trees from a fault dictionary. Method of generating candidates. By constructing the candidates from verified FMEA output, the method described in ....

Bratko, I.; Mozetic, I.; Lavrac, N. 1989. KARDIO: a study in deep and qualitative knowledge for expert systems, MIT Press.

....set of interest in terms of a subset of attributes by means of an upper and a lower bound. For example, consider the set X of patients in Table 1 who survived (p 1 ; p 5 ) and suppose we are to use the attributes age and tumour thickness in the approximation of X . The sets fp 1 ; p 3 ; p 4 g, fp 2 ; p 5 ; p 7 g and fp 6 ; p 8 g include indiscernible elements in the dataset. The B lower approximation of X (BX) where B = fage; tumour thicknessg, includes the elements of all indiscernible sets that are contained in X . In our example, only the indiscernible set fp 1 ; p 3 ; p 4 g ....

....p 3 ; p 4 g, fp 2 ; p 5 ; p 7 g and fp 6 ; p 8 g include indiscernible elements in the dataset. The B lower approximation of X (BX) where B = fage; tumour thicknessg, includes the elements of all indiscernible sets that are contained in X . In our example, only the indiscernible set fp 1 ; p 3 ; p 4 g meets this criterion. The B upper approximation of X (BX) includes the elements of all indiscernible sets whose intersection with X is nonempty, in this case this corresponds to fp 1 ; p 2 ; p 3 ; p 4 ; p 5 ; p 7 g. BX includes all elements in X for which B is sucient for their classi cation ....

[Article contains additional citation context not shown here]

I. Bratko, I. Mozetic and N. Lavrac, \KARDIO: A Study in Deep and Qualitative Knowledge for Expert Systems, The MIT Press, Cambridge, Massachusetts, 1989.

....of the learning process giving additional information about the state of the debugging to help the user answer the questions posed by the system. A detailed description can be found in [9] ffl Integrating a diagnosis module: We started integrating the diagnosis rules of the KARDIO system (see [2]) which, on the base of numerical values collected by the classifier is able to make a diagnosis for several heart diseases. The general structure of the system is displayed in Figure 6 . 4 The learning process and experimental results In the following section we briefly show how the integrated ....

Bratko I., Mozetic I., Lavrac N.: KARDIO: A study in Deep and Qualitative Knowledge for Expert Systems, The MIT Press, 1989

....air resistance) Car velocity Q I Q Gas pedal posn Car velocity I = Figure 1: The I# relation, a syntactic shorthand for a self stabilising feedback loop. models into rules #in this paradigm, there is no independent training data and the QM does not directly constrain induction e.g. #Bratko et al. 1989##, and in learning QMs themselves from examples #e.g. #Bratko et al. 1991, Mozetic, 1987##. 3 LEARNING METHOD 3.1 KNOWLEDGE REPRESENTATION Our learning method takes as input a set of training examples and a qualitative model, and as output produces classi#cation rules explainable by that ....

Bratko, I., Mozetic, I., and Lavra#c, N. #1989#. Kardio: A Study in Deep and Qualitative Knowledge for Expert Systems.

....behavior. Intermediate Depth Representations Different approaches have been explored to resolve the problems of deep and shallow models, among them multi level presentations based on variable granularity or variable presentation type [2] and transformations between knowledge representations [1]. In medicine, where the temporal aspect of a disease and treatment is very important, Coiera suggests the use of qualitative disease histories (QDHs) 2] 3] 4] In short, if the natural disease history (NDH) describes the usual temporal course of a disease in an average patient, then the ....

Bratko, I., Mozetic, I., Lavrac, N. eds. 1989. KARDIO: A Study In Deep And Qualitative Knowledge For Expert Systems. Reading, Mass.: MIT Press.

....expert systems [Pople, 1982] In this section, the logical formalization of some of these types of knowledge, and the current state of the art in research, is brie y reviewed. In Table 1, an overview of logic engineering research is listed. Logic model Application References causal simulation [Bratko et al. 1989; Lucas, 1993] diagnosis [Console et al. 1989] Torasso Console, 1989] Fox et al. 1990a; Fox et al. 1990b] anatomic design [Hammond et al. 1993] diagnosis [Lucas, 1993] taxonomic diagnosis [Huang et al. 1993] heuristic diagnosis [Fox et al. 1990b] Lucas, 1993] Moser ....

.... et al. 1989] Torasso Console, 1989] Fox et al. 1990a; Fox et al. 1990b] anatomic design [Hammond et al. 1993] diagnosis [Lucas, 1993] taxonomic diagnosis [Huang et al. 1993] heuristic diagnosis [Fox et al. 1990b] Lucas, 1993] Moser Adlassnig, 1992] functional diagnosis [Bratko et al. 1989; Coiera, 1990] monitoring [De Geus et al. 1991] treatment [De Geus et al. 1991] safety treatment [Hammond et al., 1994; Hammond Sergot, 1995] Table 1: Application of logic representation models in medicine. In applying logic to the construction of medical knowledge based systems it is ....

[Article contains additional citation context not shown here]

Bratko, I, Mozetic, I and Lavrac, N, 1989. KARDIO: A Study in Deep and Qualitative Knowledge for Expert Systems, The MIT Press, Cambridge, Massachusetts.

....to have less relevance in the construction of prognostic models than causal knowledge, which additionally may also incorporate the notion of time explicitly. A typical example of the application of a qualitative prediction model for the purpose of diagnosis has been realised in the KARDIO system [6]. Basically, KARDIO s knowledge base consists of a logical formalisation of a qualitative simulation model of the (normal and abnormal) electrical activity of the heart. The predictive model can be triggered by the assumption of the presence of a particular (combination of) cardiac arrhythmias in ....

I. Bratko, I. Mozetic and N. Lavrac, KARDIO: A Study in Deep and Qualitative Knowledge for Expert Systems (The MIT Press, Cambridge, Massachusetts, 1989).

....An initial algorithm for learning in the DHDB formalism is a part of QuMAS #Qualitative Model Acquisition System, Mozeti#c 1987#, described also by Mozeti#c and Lavra#c #1988#. It was #rst used for learning functions of components of a qualitative model of the heart in the KARDIO system #Bratko, Mozeti#c and Lavra#c 1989#. The e#ective use of background knowledge in LINUS is described by Lavra#c #1990#. The aim of this paper is to showhow LINUS can be used to learn nonrecursive logical de#nitions of relations. Section 2 introduces the DHDB formalism and gives an overview of the system, and in Section 3 the ....

Bratko, I., Mozeti#c, I. and Lavra#c, N. #1989# KARDIO: A study in deep and qualitative knowledge for expert systems. Boston, MA: The MIT Press.

....language. Keywords: Constraint Propagation, Unification, Qualitative Modelling, Implementation. 1 Introduction This paper shows how a specific constraint propagation technique namely domain variables [1] can speed up qualitative diagnosis considerably. We are using the KARDIO system [2], a qualitative simulation model of the electrical activity of the heart, to exemplify our points. Given a state of the heart (some combination of arrhythmias) the KARDIO model can be used to compute possible ECG patterns and vice versa. The design of the model influences efficiency heavily : ....

....some improvements. Section 5 discusses the results and compares our solution to other approaches. 2 The KARDIO Model The KARDIO expert system models the electrical activity of the heart in a qualitative way. We will just briefly sketch the model, an extensive description of KARDIO can be found in [2]. Overly simplified, the heart works electrically as follows: certain generators supply electrical impulses which are in turn conducted and combined through specific pathways. These resultant impulses allow the model to predict possible ECG patterns. The current version of KARDIO relates 943 ....

[Article contains additional citation context not shown here]

Bratko I., Mozetic I., Lavrac N.: Kardio - A Study in Deep and Qualitative Knowledge for Expert Systems, MIT Press, Cambridge, MA, 1989.

....way to increase the impact of machine learning might be to develop specialised conceptualising tools as Scientific Assistants. Such a tool would help scientists by suggesting interesting hypotheses from data and background knowledge, as has already been started in herbicide selection, cardiology [3] and molecular chemistry [21, 14, 39] This kind of tool certainly seems like something worth aiming for. However in Turing s vision of learning machines, the learning played a much more fundamental role. Every action involving communication between humans and computers, and even between one ....

I. Bratko, I. Mozetic, and N. Lavrac. KARDIO: a study in deep and qualitative knowledge for expert systems. MIT Press, Cambridge, 1989.

....the KBS is by describing the physiological structure of the system. This structure is used to simulate the behavior of the system. The simulation can be done either by using a model describing the physiological mechanisms in the form of qualitative relations (e.g. electrophysiology of the heart [Bratko et al. 1989, Hunter et al. 1989] or by using a qualitative, constraint model like the QSIM approach [Kuipers 1986] which is mainly based on differential equations. Qualitative simulation is well suited to model flows (of liquids, pulses, information, etc. both in the medical domain [Ironi et al. 1989] ....

Bratko I., Mozetic I., Lavrac N. (1989): Kardio - A Study in Deep and Qualitative Knowledge for Expert Systems, MIT Press, Cambridge, MA.

No context found.

BRATKO, I., MOZETI C, I., AND LAVRA C, N. 1989. KARDIO:A Study in Deep and Qualitative Knowledge for Expert Systems. MIT Press, Cambridge, MA.

....induction exhibits about the same classification accuracy with the increased transparency and lower complexity of the developed models. Michie [26] emphasized the important role of structured induction in the future and listed several real problems that had been solved in this way. Mozetic [27, 28, 7] employed another scheme for structuring the learning problem. That approach was particularly aimed at automated construction of system models from inputoutput observations of the system s behavior. The structure of the learning problem, specified by a Prolog clause, corresponded to the physical ....

I. Bratko, I. Mozetic, and N. Lavrac. KARDIO: a study in deep and qualitative knowledge for expert systems. MIT Press, 1989.

..... expert user Figure 1: An expert system schema of early 80s. 7 sets of rules [73, 66] An early approach to combining the use of deep knowledge and machine learning was used in the development of KARDIO, a system for ECG diagnosis of cardiac arrhythmias [16]. In the late eighties and early nineties it became apparent that knowledge acquired from experts alone is unsuitable for solving di#cult problems and that, when developing decision support systems, the analysis of data gathered in the daily practice of experts and stored systematically in ....

....input to the learner is a set of unclassified instances. Besides unsupervised learning using neural networks described in Section 4.2.2 and learning of association rules described in Section 3.1. 3, other forms include conceptual clustering [49, 107] subgroup discovery [172] and qualitative models [16]. The data visualization techniques may either complement or additionally support other data analysis techniques. They may be used in the preprocessing stage (e.g. initial data analysis and feature selection) and the postprocessing stage (e.g. visualization of results, tests of performance of ....

Bratko, I., Mozetic, I. and Lavrac, N., KARDIO: A Study in Deep and Qualitative Knowledge for Expert Systems, The MIT Press, 1989.

....and relational descriptions. The LINUS algorithm is a descendant of the learning algorithm used in QuMAS (Qualitative Model Acquisition System) which was used to learn functions of components of a qualitative model of the heart in the KARDIO expert system for diagnosing cardiac arrhythmias [3]. The method, implemented in the system LINUS, employs propositional learners in a more expressive logic programming framework. 9 3.1 Learning with LINUS LINUS is an ILP learner which induces hypotheses in the form of constrained deductive hierarchical database (DHDB) clauses (a formal ....

I. Bratko, I. Mozetic, and N. Lavrac. KARDIO: A Study in Deep and Qualitative Knowledge for Expert Systems. MIT Press, Cambridge, MA, 1989.

....induction exhibits about the same classification accuracy with the increased transparency and lower complexity of the developed models. Michie [25] emphasized the important role of structured induction in the future and listed several real problems that had been solved in this way. Mozetic [26, 27, 7] employed another scheme for structuring the learning problem. That approach was particularly aimed at automated construction of system models from inputoutput observations of the system s behavior. The structure of the learning problem, specified by a Prolog clause, corresponded to the physical ....

I. Bratko, I. Mozetic, and N. Lavrac. KARDIO: a study in deep and qualitative knowledge for expert systems. MIT Press, 1989.

....reported by Genesereth (1984) and Korf (1987, in planning) but without any experimental evidence to support the claim. An application of abstractions to a complex medical problem, described in section 5, confirms the expected complexity reduction. The problem originating from the KARDIO project (Bratko, Mozetic Lavrac 1989) is to diagnose heart disorders, given an ECG and a simulation model of the heart s electrical activity. Until now, all attempts to directly use the model for efficient diagnosis failed in an average case more than 50sec: is needed to find all diagnoses. By abstractions and refinements, the ....

....fact that, while the total number of states grows exponentially, the number of states to be verified is kept constant across levels. In our experiments this actually turned out to be the case. 5 Experiments and results The experimental evaluation involves a realistic medical problem. In KARDIO (Bratko, Mozetic Lavrac 1989), the ECG interpretation problem is formulated as follows: given a symbolic ECG description ECG, find all possible single and multiple heart disorders (cardiac arrhythmias Arr) In the medical literature there is no systematic description of ECG features which correspond to complicated ....

Bratko, I., Mozetic, I., Lavrac, N. (1989). KARDIO: A Study in Deep and Qualitative Knowledge for Expert Systems. The MIT Press, Cambridge.

.... in terms of constraints coupled with an ATMS [6, 7] or as a set of propositions in first order logic [18] In contrast, we represent models by logic programs [11] and by constraint logic programs [10] Similar representation was proposed in [19] but the origin goes back to the KARDIO model [1]. Definitions of basic concepts typically follow [18] we give an alternative, relational characterization, suitable for model representation by (constraint) logic programs. Definition. A model of a system is a triple hSD, COMPS, OBSi where: 1. SD, the system description, is a (constraint) ....

....methods used by ATMSbased systems. CLP( was applied to diagnose analog circuits operating under the AC conditions [14] It enables modeling of soft faults drifts from the nominal parameter values, and computation with parameter tolerances. In a medical application, the heart model in KARDIO [1] was specified by a pure logic program, and recently reformulated in terms of constraints over finite domains [15] IDA works with any of the above models. Acknowledgements This work was supported by the Austrian Federal Ministry of Science and Research. Thanks to Christian Holzbaur for his ....

Bratko, I., Mozetic, I., Lavrac, N. KARDIO: A Study in Deep and Qualitative Knowledge for Expert Systems. MIT Press, Cambridge, 1989.

.... to the investigations of the use of deep causal knowledge that could potentially overcome the difficulties encountered when using unstructured shallow level sets of rules [ 20; 18 ] An early approach in combining the use of deep knowledge and machine learning is presented in the KARDIO study [ 5 ] In late eighties and early nineties it thus became apparent that knowledge acquired from experts alone is unsuitable for solving difficult problems and that, when developing decision support systems, the analysis of data gathered in the daily practice of experts and stored systematically in ....

Bratko, I., Mozetic, I., and Lavrac, N. (1989). KARDIO: A Study in Deep and Qualitative Knowledge for Expert Systems. MIT Press, Cambridge, MA.

No context found.

I. Bratko, I. Mozetic, and N. Lavrac. KARDIO: a Study in Deep and Qualitative Knowledge for Expert Systems. MIT Press, 1989.

No context found.

Bratko, I., I. Mozetic, and N. Lavrac, KARDIO: a study in deep and qualitative knowledge for expert systems. 1989, Cambridge, MA: The MIT Press. 260.

No context found.

I. Bratko, I. Mozetic, N. Lavrac, "KARDIO: A Study in Deep and Qualitative Knowledge for Expert Systems", Cambridge, Massachusetts: MIT Press, 1989. 7

No context found.

I. Bratko, I. Mozetic, and N. Lavrac, KARDIO: A Study in Deep and Qualitative Knowledge for Expert Systems, MIT Press, London, England, 1989.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC