T.J. Menzies and P. Compton. Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative Compartmental Models. Artificial Intelligence in Medicine, 10:145-175, 1997. Available from h;;p://wm. cse .unsw. edu. au/~;in/pub/ docs/96aim.ps. gz.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of temporal reasoning methods for QCM. On the basis of this a general methodology for evaluating QRSs and a statement of success criteria have been developed, and will be used in future work. 1 Introduction This paper examines a number of methods for extending an existing QRS, the QCM algorithm [10], to handle time based reasoning. This system contains a static model compiler, a data compiler, and a hypothesis tester for the model and data. The semantics of QCM are extended to allow time series data, giving QCM T . Six linking policies are described and tested in this work to implement ....

....tester for the model and data. The semantics of QCM are extended to allow time series data, giving QCM T . Six linking policies are described and tested in this work to implement temporal reasoning. Overview articles which contrast different approaches to qualitative reasoning (QR) e.g. [2 4, 6, 10, 11]) have little to say about how to choose between different systems. There is also a lack of guidelines for developing and testing new modelling approaches. In the process of testing the different QCM T linking methods we have developed a more general framework for assessing QR systems. We ....

[Article contains additional citation context not shown here]

T.J. Menzies and P. Compton. Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative CompartmentalModels. Artificial Intelligence in Medicine, 1997. To appear.

....what are reasonable lower bounds on the size of a test suite. Size will be expressed as the percentage of variables in a theory which are not measured in the test data set. The lower bound on size will be found as follows. Assuming our temporal graph theoretic abductive validation procedure [8, 11, 12, 17], we will reduce test suite size until we can no longer distinguish good theories from bad theories. It will be found that seemingly minor variations in a language can have a significant impact on this lower bound on test suite size. This paper is hence a cautionary note to those who invent ....

....dependent on minor variations in the language used. For example, XNODE is only a small variant on the lEDGE language. However, XNODE is practical down to 70 percent unmeasured while lEDGE fails after 40 percent unmeasured. The theoretical framework of this article has been presented before (e.g. [8, 11, 12, 17]) The new contribution of this article is the data reduction ex periments and the observation that languages react very differently to data reduction. 2 Theories This section describes the types of theories used in this analysis. The next section describes a validation procedure which executes ....

[Article contains additional citation context not shown here]

T.J. Menzies and P. Compton. Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative Compartmental Models. Artificial Intelligence in Medicine, 10:145-175, 1997. Available from h;;p://wm. cse .unsw. edu. au/~;in/pub/ docs/96aim.ps. gz.

....have ( D p o q r ) arguments as they discuss which portions of changes1 to apply. With TARZAN, our stakeholders need only have ( o (p ) arguments. Reasoning in the presence of uncertainty: Uncertainty is fundamental to many under measured domains, e.g. human internal medicine, economics [13], and software development. For example, software development data may be scarce if the development team was not funded to maintain a metrics repository, or the collected data does not relate to the business case, or if contractors prefer to retain control of their own information. After two years ....

....by management directives to collect more data. Our view is that we should take data famines as a premise, and then research how to reason in their presence. Our prior research into reasoning about uncertain domains used abductive logics to implement HT4: a set covering truth maintenance system [13]. Such logical approaches to uncertain reasoning can be too slow: HT4 s inference is provably NP hard and experimentally exponential on model size. TARZAN, on the other hand, is a much simpler, faster method. TARZAN learns emergent stable properties from a wide range of randomly selected behavior. ....

T.J. Menzies and P. Compton. Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative Compartmental Models. Artificial Intelligence in Medicine, 10:145-- 175, 1997. Available from http://www.cse.unsw.edu.au/ timm/pub/docs/96aim.ps.gz.

No context found.

T.J. Menzies and P. Compton, `Applications of abduction: Hypothesis testing of neuroendocrinological qualitative compartmental models', Artificial Intelligence in Medicine (1997). To appear.

....= z # # ands # pands = y # # pands # pors = y # # pors y =0. 0.5 z = 0.5 . 2.0 repeats = 1. 20 Table 1: Simulation parameters Simulations Table 1 shows parameters for an and or graph from the real world model of neuroendocrinology (Smythe 1989) explored previously by Menzies Compton (Menzies Compton 1997). Initial experiments with our formulae assumed y = 0; repeats = 1; i.e. the number of parents of and nodes and or nodes was constant. The results of the y =0; repeats =1run are shown in Figure 2. Note that for a range of graph sizes and number of constraints, the same pattern emerges: # ....

Menzies, T., and Compton, P. 1997. Applications of abduction: Hypothesis testing of neuroendocrinological qualitative compartmental models. Artificial Intelligence in Medicine 10:145--175. Available from http://www.cse.unsw.edu.au/timm/ pub/docs/96aim.ps.gz.

....ideas and tolerate inconsistencies in an evolving specification. Further, QCM has a built in validation engine which can help experts to review and improve their current specification. Hence, we say that QCM is a conceptual modeling language [Feldman et al. 1989a, Menzies, 1995, Menzies, 1996a, Menzies Compton, 1997,Waugh et al. 1997] We will use the criteria of KB testability and KBmaintainability (defined below) to demonstrate the need for certain restrictions to QCM conceptual models. Roughly speaking, KB testability means checking if a theory of X can reproduce known behaviour of X; and ....

....clauses, or qualitative theories used for simulations. Further, the methods used here (instance generators, graph theory, analysis of KB testability and KB maintainability) are quite general and could be used to find restrictions to other conceptual modeling languages. We have argued elsewhere [Menzies, 1997] that the KE field urgently needs such general evaluation methodologies. Our conclusions come from a theoretical and empirical analysis of a KBtestability and KB maintainability algorithm. We take care to distinguish this work from the worst case time complexity research of the knowledge ....

[Article contains additional citation context not shown here]

Menzies, T. & Compton, P. (1997). Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative Compartmental Models. Artificial Intelligence in Medicine, 10:145--175.

....what are reasonable lower bounds on the size of a test suite. Size will be expressed as the percentage of variables in a theory which are not measured in the test data set. The lower bound on size will be found as follows. Assuming our temporal graph theoretic abductive validation procedure [8, 11, 12, 17], we will reduce test suite size until we can no longer distinguish good theories from bad theories. It will be found that seemingly minor variations in a language can have a significant impact on this lower bound on test suite size. This paper is hence a cautionary note to those who invent ....

....dependent on minor variations in the language used. For example, XNODE is only a small variant on the IEDGE language. However, XNODE is practical down to 70 percent unmeasured while IEDGE fails after 40 percent unmeasured. The theoretical framework of this article has been presented before (e.g. [8, 11, 12, 17]) The new contribution of this article is the data reduction experiments and the observation that languages react very differently to data reduction. 2 Theories This section describes the types of theories used in this analysis. The next section describes a validation procedure which executes ....

[Article contains additional citation context not shown here]

T.J. Menzies and P. Compton. Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative Compartmental Models. Artificial Intelligence in Medicine, 10:145--175, 1997. Available from http://www.cse.unsw.edu.au/~timm/pub/ docs/96aim.ps.gz.

....search spaces are less complex than Earnest thinks. Hence, the test suites required to sample that search space may be very small. However, the sample size of the above studies is very small. This section addresses the sample size issue. Based on the HTx abductive model of testing (Menzies 1995; Menzies Compton 1997), a suite of mutators can generate any number of sample testing problems 4 . Three 4 Informally, abduction is inference to the best explanation. More precisely, abduction make whatever assumptions A which are required to reach output goals Out across a theory T [ A Out without causing ....

....no other comment on the nature of Out i . It may be some undesirable state or some desired goal. In either case, the aim of our testing is to find the worlds that cover the largest percentage of Out. An Example Figure 3 (left) shows a hypothetical economics theory written in the QCM language (Menzies Compton 1997). All theory variables have three mutually exclusive states: up, down or steady; i.e. no(V a=up ) fV a=down ; V a=steady g and constraints(G) 2. These values model the sign of the first derivative of these variables (i.e. the rate of change in each value) In QCM, x y denotes that y ....

[Article contains additional citation context not shown here]

Menzies, T., and Compton, P. 1997. Applications of abduction: Hypothesis testing of neuroendocrinological qualitative compartmental models. Artificial Intelligence in Medicine 10:145--175. Available from http://www.cse.unsw.edu.au/timm/ pub/docs/96aim.ps.gz.

....we use in this paper for exploring multiple worlds reasoning is graph based abduction. Informally, abduction is the inference to the best explanation [24] More precisely, abduction makes assumptions in order to complete some inference. Mutually exclusive assumptions are managed in separate worlds [19]. That is, given a theory containing contradictions, abduction sorts those contradictions into consistent portions. In this case, the theory is the union of the viewpoints of different stakeholders. Queries can be written to assess the different worlds. Abduction allows us to examine the ....

....Dr. Thick s and Dr. Thin s ideas are shown in thick and thin lines respectively. Squares denote and nodes Colin, our requirements engineer, has interviewed two software managers, Dr. Thick and Dr. Thin, to create the viewpoints of figure 1. These viewpoints are recorded using the QCM notation [19]. Squares denote and nodes: i.e. conjunctions that follow some premise. Each variable has three states: up, down or steady. These values model the sign of the first derivative of these variables. There are two types of dependencies between them, as follows: Dr. Thin s direct connection between ....

[Article contains additional citation context not shown here]

T. Menzies and P. Compton. Applications of abduction: Hypothesis testing of neuroendocrinological qualitative compartmental models. Artificial Intelligence in Medicine, 10:145--175, 1997.

....are at least three ways to check an expert s behaviouralK: 1. The behaviouralK could reflect records of the known behaviour of the domain being modeled. However, in practice, such real world data sets are rare. Menzies Compton have argued that many domains tacked by 860 KBS are very data poor [73]. 2. A new KB2 could be used (and this would introduce the maintenance recursion problem) This KB2 could some how detect if the behaviouralK does not cover a representative sample of the domain. 3. If a test coverage tool reports that behaviouralK fails to exercise an ade 865 quate portion of ....

T.J. Menzies and P. Compton. Applications of Abduction: Hypothesis Testing of 1530 Neuroendocrinological Qualitative Compartmental Models. Artificial Intelligence in Medicine, (10):145--175, 1997.

....This framework is based around the HT4 graph theoretic abduction engine (x2.1) HT4 searches for consistent portions of some background theory which are relevant to some task. The approach to abduction was originally defined for the valida 75 tion of qualitative neuroendocrinological theories [16] (x3.7) Once developed, it was noted that the internal data structures of the algorithm made little commitment to qualitative reasoning. Indeed, the algorithm could execute over any representation reduced to a directed graph of literals; e.g. business process graphs (x2.2) connecting objects ....

....67 . The maximum cover is 100 ; i.e. i) their exist a set of assumptions (fcUpg) which let us explain all of OUT ; and (ii) this model has based abductive validation. 340 Note that this process can be summarised as: can a model of X explain known behaviour of X . We have argued elsewhere [16] that this is the nonnaive implementation of KBS validation since it handles certain interesting cases: If a model is globally inconsistent, but contains local portions that are consistent and useful for explaining some behaviour, HT4 will find those por 345 tions. In the situation ....

[Article contains additional citation context not shown here]

T.J. Menzies and P. Compton. Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative Compartmental Models. Artificial Intelligence in Medicine, 1997. To appear.

....language can have an enormous impact of the computational complexity of inference over that representation. In this paper we explore the testability of a temporal extension to the LQCM language (x3) Theories expressed in LQCM can be tested using our HT4 abductive validation engine [Men96b, MC97]. While abductive validation approach has proved useful for an interesting range of real world theories (x2) it cannot be extended to LTQCM : i.e. temporal abductive validation which tests theories used in a time based simulation (x4) the differences between abductive validation and temporal ....

....eUp, fDowng B. D1 computed from T1 bDown aUp xUp eUp fUp gUp cUp dUp yUp yDown eDown fDown gDown cDown bUp xDown fSteady 001 004 003 005 dSteady dDown 006 007 008 aDown xSteady cSteady 002 Fig. 1. Theories to dependency graphs 3 LQCM LQCM is a qualitative compartmentalmodeling language [MC97]. Compartmental models include tubs with in flows and out flows. In quantitative compartmental modeling, the current level of liquid in a tub equals its initial value plus its inflows, minus its out flows. In qualitative compartmental modeling, the sign of the the rate of change (first derivative) ....

[Article contains additional citation context not shown here]

T.J. Menzies and P. Compton. Applications of abduction: Hypothesis testing of neuroendocrinological qualitative compartmental models. Artificial Intelligence in Medicine, 1997. To appear.

.... systems that interact directly with the environment without first reflecting over some symbolic description [73] While we have some sympathies to the weak SC line (see x2.1) we hope that 40 strong SC is an over reaction since we are great fans of symbolic approaches to building KBs (e.g. [76 78]) We would like to demonstrate that strong SC is an over reaction via a review of the knowledge maintenance (KM) literature. SC and KM are intimately connected: ffl SC motivates KM: If knowledge is situated, then a symbolic KB will require 45 modification as the situation changes. ffl KM ....

....expect specifications to reflect different and inconsistent viewpoints. ffl Feldman Compton reports that, in terms of being able to reproduce 125 known behaviour, certain theories published in the international peerreviewed literature are grossly inaccurate [39] a result repeated by Menzies [78]) ffl Kuhn notes that data is not interpreted neutrally, but (in the usual case) processed in terms of some dominant intellectual paradigm [60] 130 ffl Explicit records of domain knowledge may not evolve to some final stable point, even when the domain itself was stable. Compton reports one ....

[Article contains additional citation context not shown here]

T.J. Menzies and P. Compton. Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative Compartmental Models. Artificial Intelligence in Medicine, 1997. To appear.

....search for consistent subsets of some background theory that are relevant for achieving some goal. If multiple such subsets can be generated, then a BEST assessment operator selects the preferred subset(s) or worlds, W) For example, consider our graphtheoretic HT4 abductive inference engine [33,34,36]. HT4 determines what OUTput goals can be reached from using the INputs shown in a knowledge base like Figure 1. In that Figure, x y denotes that y being up or down can be explained by x being up or down respectively and x Gamma Gamma y denotes that y being up or down could be ....

....question: how much of the known behaviour of X can be reproduced by out model of X . This abductive validation procedure has faulted theories published in the international peer reviewed literature. Interesting, we have found these faults using the data published to support those theories [36]. We have argued elsewhere [33,34] that this is the non naive implementation of KBS validation since it handles three certain interesting cases: 1. If not all variables in the theory are measured, then this procedure can still validate the theory. The procedure will take the necessary assumptions ....

T.J. Menzies and P. Compton. Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative Compartmental Models. Artificial Intelligence in Medicine, 1997. To appear.

....and consensus. The focus of this paper is on consensus, not on the evaluation of the competency of the individual or combined rock classification KBS. Such an evaluation is problematic as precise evaluation standards have not been defined. Automatic methods of assessing competency are explored in Menzies and Compton (1997). This paper is organised as follows. Section 2 introduces RE, Section 3 describes our RE framework and our implementation in found in section 4. A case study is presented in Section 5 using seven knowledge bases built from the data for the SISYPHUS III project. Our evaluation technique is ....

....modification to the use cases. Prior work in this area has been presented as verification or maintenance research (Compton et al. 1992, Richards and Compton 1997a) We here claim that an RE technique could be viewed as a technique for knowledge maintenance (KM) validation and verification (V V) Menzies (1997) argues that knowledge management techniques amount to a small number of activities processing less than a dozen types of 18 knowledge. At different points of the software lifecycle, the emphasis is on certain activities processing some of the knowledge types: At later stages of the life ....

Menzies, T. And Compton, P. (1997) Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative Compartmental Models, Artificial Intelligence in Medicine, 10:145-175.

....models of three geology experts about some standard 40 problem in their field. Feldman Compton reports that, in terms of being able to reproduce known behaviour, certain theories published in the international peer reviewed literature are grossly inaccurate [7] a result repeated by Menzies [15]) Anderson reports carefully constructed experiments showing people failing to see as valid certain conclusions that are valid, and they see as valid 45 certain conclusions that are not. 2, p264] Kuhn notes that data is not interpreted neutrally, but (in the usual case) processed in terms of ....

....run graph i as follows. Randomly select up to jIN j vertices to be INputs and up to jOUT j different vertices to be OUT puts. Call HT4 with those INputs and OUT puts. For these experiments, graph 0 was the and or graph of the Smythe 89 theory of human glucose regulation [18] described in [15]) For that graph: 195 1. jVj = 554 2. jE j jV j = 2:25 3. 1 jIN j 4 4. 1 jOUT j 10 Figure 4 was generated by mutator 1 which increased jVj while keeping jIN j 200 jOUT j and jE j jVj constant [12,13] The results below were generated by mutator 2 which which kept jVj, jIN j and jOUT j ....

T.J. Menzies and P. Compton. Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative CompartmentalModels. Artificial Intelligence in Medicine, 350 1997. To appear.

No context found.

T.J. Menzies and P. Compton, `Applications of abduction: Hypothesis testing of neuroendocrinological qualitative compartmental models', Artificial Intelligence in Medicine (1997). To appear.

....studied here is only a flat causal structure without terms in a subsumption hierarchy. However, our belief (as yet untested) is that the methods used here (mutators, analysis of KB maintainability) are quite general and could be used to find restrictions to other ontologies. Menzies has argued [Menzies, 1997] that the KE field needs such general evaluation methodologies. This paper is structured as follows. Our ontology was developed in response to certain drawbacks with the 1980s research into qualitative reasoning. That work is reviewed first, followed by our ontology. Next, we explore how we would ....

....defined to assess these options. When the experiment was performed, it was found that not all these options were useful. Our discussion section generalises this work into a framework for assessing ontologies. 2 Causality, Qualitative Reasoning, and QCM Qualitative compartmental modeling (QCM) [Menzies Compton, 1997,Waugh et al. 1997] offers a physical causal ontology for testing hypotheses in neuroendocrinology (the study of nerves and glands) QCM is a generalisation of QMOD: Feldman and Compton s work [Feldman et al. 1989a, Feldman et al. 1989b] on qualitative reasoning (QR) This section reviews the ....

[Article contains additional citation context not shown here]

Menzies, T. & Compton, P. (1997). Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative Compartmental Models. Artificial Intelligence in Medicine, 10:145--175.

....in expert preferences may result in incorrect incomplete knowledge bases. Feldman 90 Compton reports that, in terms of being able to reproduce known behaviour, certain theories published in the international peer reviewed literature are grossly inaccurate [43] a result repeated by Menzies [70]) Preece Shinghal [78] document five fielded expert systems that contain numerous logical anomalies (see Figure 1) These expert systems still work, apparently because in the context of 95 their day to day use, the anomalous logic is never exercised. Shaw reports an experiment where a group ....

....SC is arguing that using truth changes truth . We can offer a concrete example of exactly this process. The QMOD HT4 project built automatic tools to faults in neuroendocrinological theories of glucose 245 regulation (prototype described in [42] generalised and formalised version described in [70], discussed here in x4.3) and reported: Menzies: KM SC; page 8 of 28 0 10 20 30 40 0 10 20 30 40 50 Kilobytes Months in maintenance Observed Theta Theta Theta Theta Theta Theta Theta Theta Linear fit; r 2 = 0:91 Logarithmic fit;r 2 = 0:89 Figure 5: Garvan ES 1 knowledge base size ....

[Article contains additional citation context not shown here]

T.J. Menzies and P. Compton. Applications of Abduction: Hypothesis Testing of Neur- 900 oendocrinological Qualitative Compartmental Models. Artificial Intelligence in Medicine, 1997. To appear.

....experience with the earlier versions of the DR C algorithm suggested that the runtimes grew alarmingly as the size of D T increased. An alternative to chronological backtracking is an algorithm that caches what it learns about the search space as it executes. Our HT4 abductive inference algorithm [51, 52, 55] is a such a cachebased search algorithm that implements DR C. HT4 caches the proof trees used to satisfy EQ 1 and EQ 2 . These are then sorted into worlds using the base controversial assumptions AB. These procedure worldsSweep begin ENV : maximalConsistentSubsets(AB) for i : 1 to ....

T.J. Menzies and P. Compton. Applications of Abduction: Hypothesis Testing of Neuroendocrinological Qualitative Compartmental Models. Artificial Intelligence in Medicine, 1997. To appear.

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