#### DMCA

## Learning Stochastic Logic Programs (2000)

### Cached

### Download Links

Citations: | 1173 - 79 self |

### Citations

6460 |
C4.5: Programs for Machine Learning.
- Quinlan
- 1993
(Show Context)
Citation Context ...learning have been successfully applied to various problems [1, 12]. Most of these applications rely on attribute-based learning, exemplified by the induction of decision trees as in the program C4.5 =-=[20]-=-. Broadly speaking, attribute-based learning also includes such approaches to learning as neural networks and nearest neighbor techniques. The advantages of attribute-based learning are: relative simp... |

2070 |
Foundations of Logic Programming
- Lloyd
- 1987
(Show Context)
Citation Context ... used programming language [3, 41] and was chosen as thescore language for Japan's Fifth Generation Project. Prolog's development has alsosspawned the rigorous theoretical school of Logic Programming =-=[20]-=-.s1.2 InductionsDespite the self-evident success of logical deduction, a certain question has croppedsup time and again throughout i s development. If all human and computer ea-ssoning proceeds from l... |

1960 | A theory of the learnable - Valiant - 1984 |

1102 |
Language identification in the limit.
- Gold
- 1967
(Show Context)
Citation Context ... clause is faulty and thensreplaces that clause.sBoth Plotkin and Shapiro managed to prove that their learning systems couldsidentify first order theories in the limit, according to Gold's definition =-=[13]-=-. Nowa-sdays Valiant's definition of PAC-learning [14] (see Section 3.4) is generally agreedsto provide a better approach to identification than Gold's methodology. However,sPAC methods have largely b... |

1092 |
A Machine-oriented Logic Based on the Resolution Principle.
- Robinson
- 1965
(Show Context)
Citation Context ...Ghdel's demonstration [9]that a smallscollection of sound rules of inference was complete for deriving all consequencessof formulae in first order predicate calculus. Much later Robinson demonstrateds=-=[34]-=- that a single rule of inference, called resolution, is both sound and completesfor proving statements within this calculus (see Appendix A). For the purposessof applying resolution, formulae are norm... |

927 | Learning logical definitions from relations
- Quinlan
- 1990
(Show Context)
Citation Context ...cle.sAn attempt to overcome the limitations of attribute-based learning has led to recent development of a number of programs that learn at the level of firstorder predicate logic. These include FOIL =-=[19]-=-, Golem [17], and Progol [22], with Shapiro’s program MIS as one of their early predecessors [21]. This has led to the inception of a new area of machine learning called Inductive Logic Programming [1... |

715 | Inverse entailment and Progol - Muggleton - 1995 |

580 | Explanation-based generalization: A unifying view - Mitchell, Keller, et al. - 1986 |

547 | Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme. Monatshefte für Mathematik und Physik - Gödel - 1931 |

537 | A note on inductive generalization - Plotkin - 1970 |

529 |
Algorithmic Program DeBugging
- Shapiro
- 1983
(Show Context)
Citation Context ...ment of a number of programs that learn at the level of firstorder predicate logic. These include FOIL [19], Golem [17], and Progol [22], with Shapiro’s program MIS as one of their early predecessors =-=[21]-=-. This has led to the inception of a new area of machine learning called Inductive Logic Programming [13]. For recent developments see [14]. ILP benefits from the solid theoretical framework provided ... |

517 |
A formal theory of inductive inference
- Solomonoff
- 1964
(Show Context)
Citation Context ...be it propositional logic, first-order logic or context-free grammar ules.sThirdly, this would bring the PAC-learning model into line with other approachessdeveloped from algorithmic omplexity theory =-=[40, 6, 2]-=- and Bayesian statistics.sAccording to Bayes' LawsPr(HIO ) = Pr(H)'Pr(OIH)sPr(O)sIf we treat bit-encoded escriptions as though they were the outcome of repeat-sedly tossing an unbiased coin, the prior... |

512 | Qualitative simulation - Kuipers - 1986 |

437 |
Logic for Problem Solving
- Kowalski
- 1979
(Show Context)
Citation Context ...ormalised to what is known as clausal form.sRobinson's discovery was of fundamental importance to the application of logicswithin computer science. Throughout the early 1970's Colmerauer and Kowalskis=-=[17]-=- were instrumental in the development of the logic based programming lan-sguage Prolog. Prolog statements are phrased in a restricted clausal form calledsHorn clause logic. All computations within Pro... |

424 | The estimation of stochastic context-free grammars using the inside-outside algorithm. Computer Speech and Language - Lari, Young - 1990 |

296 | Towards a mathematical theory of inductive inference - Blum, Blum - 1975 |

252 |
Quantifying inductive bias: AI learning algorithms and valiant’s learning framework
- Haussler
- 1988
(Show Context)
Citation Context ...Plotkin and Shapiro managed to prove that their learning systems couldsidentify first order theories in the limit, according to Gold's definition [13]. Nowa-sdays Valiant's definition of PAC-learning =-=[14]-=- (see Section 3.4) is generally agreedsto provide a better approach to identification than Gold's methodology. However,sPAC methods have largely been applied to propositional-level l arning and have a... |

226 |
The Art of Prolog: Advanced Programming Techniques
- Shapiro, Sterling
- 1994
(Show Context)
Citation Context ...ic. All computations within Prolog take the form of logical proofssbased on the application of the rule of resolution. Since its inception Prolog hassdeveloped into a widely used programming language =-=[3, 41]-=- and was chosen as thescore language for Japan's Fifth Generation Project. Prolog's development has alsosspawned the rigorous theoretical school of Logic Programming [20].s1.2 InductionsDespite the se... |

212 |
A general lower bound on the number of examples needed for learning
- Ehrenfeucht, Haussler, et al.
- 1989
(Show Context)
Citation Context ...s itswould seem likely that the area will subdivide into three related strands of research:stheory, implementation a d experimental pplication. While model-theoretic se-smantics [20] and PAC-learning =-=[8]-=- theory are likely to be influential, neither seemstotally adequate for the problems involved. The general goals of research for suchsan area should be to produce a widely used technology with a firm ... |

201 |
Buntine: Machine Invention of First-Order Predicates by Inverting Resolution
- Muggleton, W
- 1988
(Show Context)
Citation Context ...ve.s9 Strong bias ofsvocabulary. Present inductive systems construct hypothe-sses within the limits of a fixed vocabulary of propositional ttributes. Ansincreasing amount of Machine Learning research =-=[24, 28, 1, 44, 35, 16, 19]-=-sis concerned with algorithms capable of inventing auxiliary predicates whensinsufficient background knowledge is provided.s2 Inductive Logic ProgrammingsA growing body of researchers have started to ... |

196 |
Prolog programming for Artificial Intelligence
- BRATKO
- 1986
(Show Context)
Citation Context ...ic. All computations within Prolog take the form of logical proofssbased on the application of the rule of resolution. Since its inception Prolog hassdeveloped into a widely used programming language =-=[3, 41]-=- and was chosen as thescore language for Japan's Fifth Generation Project. Prolog's development has alsosspawned the rigorous theoretical school of Logic Programming [20].s1.2 InductionsDespite the se... |

196 | A theory of clausal discovery
- Raedt, Bruynooghe
- 1993
(Show Context)
Citation Context ...form of closed-world assumption which gives rise to several thousands of negative examples.sSeveral relational learning algorithms were tried on this data including Golem [17], FOIL [19] and CLAUDIEN =-=[6]-=-. The resulting set of rules were of interest to expert users of the finite element methods. According to their comments, these rules reveal interesting relational dependences that the experts had not... |

168 | Generating production rules from decision trees - Quinlan - 1987 |

167 |
Systems of logic based on ordinals
- Turing
- 1939
(Show Context)
Citation Context ...: Expert systems built with and without use of inductionsself-contradictory or incomplete for the purposes of deriving certain arithmeticsstatements. This discovery prompted Turing to attempt to show =-=[42]-=- that problemssconcerning incompleteness of logical theories could be overcome by the use of ansoracle capable of verifying underivable statements. Turing [43] later came to believesthat GSdel's incom... |

131 | A model of inexact reasoning in medicine - Shortliffe, Buchanan - 1975 |

124 |
The Continuum of Inductive Methods
- Carnap
- 1952
(Show Context)
Citation Context ...statements. Turing [43] later came to believesthat GSdel's incompleteness theorem required that intelligent machines be capablesof learning from examples.sVarious logical positivists including Carnap =-=[5]-=- have developed statistical theo-sries for confirming scientific hypotheses posed in first order logic. Although Plotkins[31] in the 1970's and Shapiro [37] in the 1980's worked on computer-based indu... |

122 | Relational Bayesian networks
- Jaeger
- 1997
(Show Context)
Citation Context ...c grammars (Lari & Young 1990) and Hidden Markov Models (HMMs)) and b) declarative representations of uncertain statements (eg. probabilistic logics (Fagin & Halpern 1989) and Relational Bayes’ nets (=-=Jaeger 1997-=-)). Stochastic Logic Programs (SLPs) (Muggleton 1996) were introduced originally a way of lifting stochastic grammars (type a representations) to the level of first-order Logic Programs (LPs). Later C... |

116 | Applications of Machine Learning and Rule Induction
- Langley, Simon
- 1995
(Show Context)
Citation Context ...orms as general as Prolog programs. This often completely changes the nature of representational engineering, which is essential part of machine learning applications as observed by Langley and Simon =-=[12]-=-. On the other hand, a major obstacle to more effective use of ILP at present is the relative inefficiency of the existing ILP systems, and their rather limited facilities for handling numerical data.... |

114 |
Drug design by machine learning: The use of inductive logic programming to model the sturcture-activity relationships of trimethoprim analogues binding to dihydrofolate reductase
- King, Muggleton, et al.
- 1992
(Show Context)
Citation Context ...P applications involved applying Golem [17] to protein secondary structure prediction [14], prediction of ß-sheet topology and modeling the structure activity relationship (QSAR) of a series of drugs =-=[11]-=-. For secondary structure prediction Golem yielded predictive accuracies well in excess of any other contemporary approach. In the case of QSAR, predictive accuracies were not significantly higher tha... |

109 | Mutagenesis: Ilp experiments in a non-determinate biological domain
- Srinivasan, Muggleton, et al.
- 1994
(Show Context)
Citation Context ...he limitations of attribute-based learning has led to recent development of a number of programs that learn at the level of firstorder predicate logic. These include FOIL [19], Golem [17], and Progol =-=[22]-=-, with Shapiro’s program MIS as one of their early predecessors [21]. This has led to the inception of a new area of machine learning called Inductive Logic Programming [13]. For recent developments s... |

108 | Protein secondary structure prediction using logic-based machine learning - Muggleton, King, et al. - 1992 |

106 | Learning Concepts By Asking Questions
- Sammut, Banerji
- 1986
(Show Context)
Citation Context ...at despite the identification i the limit guaranteessfor Plotkin's and Shapiro's ystems, both were very inefficient and were in practicesonly demonstrated on very limited problems.sSammut and Banerji =-=[36]-=- describe a system called MARVIN which generalises assingle example at a time with reference to a set of background clauses. At each stagesa set of ground atoms F representing the example are matched ... |

100 |
Generalized subsumption and its applications to induction and redundancy
- Buntine
- 1988
(Show Context)
Citation Context ...ral extension of ID3 [32]. He notessthat the search can be highly myopic, and is unable to learn predicates uch asslist-reversal nd integer-multiplication.sAttempts have been made recently by Buntine =-=[4]-=-, Frisch and Page [11] andsMuggleton and Feng [29] to find ways around Plotkin's negative RLGG results.s300 S. MuggletonsDomain Clause No. of No. of atomssexamples in backgrounds169 5408 Qualitativesm... |

84 |
Automatic Methods of Inductive Inference
- Plotkin
- 1971
(Show Context)
Citation Context ...blesof learning from examples.sVarious logical positivists including Carnap [5] have developed statistical theo-sries for confirming scientific hypotheses posed in first order logic. Although Plotkins=-=[31]-=- in the 1970's and Shapiro [37] in the 1980's worked on computer-based induc-stive systems within the framework of full first order logic, most successes within thesfield of Machine Learning have deri... |

83 |
Equivalences of logic programs
- Maher
- 1988
(Show Context)
Citation Context ...ing theoremsproving. AlsosDefinit ion 5 Literal l is logically redundant within the clause C V l in the logicsprogram P A (C V l) if and only if P A (C V l) is logically equivalent to P A C.sMaher in =-=[21]-=- discusses a number of different notions of equivalence for logic pro-sgrams including that of Definition 3. See also Buntine [4] and Niblett [30] for asdetailed iscussion of generality.s3 .2sA genera... |

82 |
A Model of Inexact Reasoning
- Shortliffe, Buchanan
- 1975
(Show Context)
Citation Context ...ional logic. The major successes here have been in thesarea of inductive construction ofexpert systems (see [27]). The properties of varioussexpert systems are given in Figure 1. The first two, MYCIN =-=[38]-=- and XCON [10]swere built using hand-coding of rules. The second two, GASOIL [39] and BMT [15]swere built using software derived from Quinlan's inductive decision tree buildingsalgorithm ID3 [32]. It ... |

81 | Deraedt L. Inductive logic programming — theory and methods - Muggleton - 1994 |

73 | Logical depth and physical complexity
- Bennett
- 1988
(Show Context)
Citation Context ...be it propositional logic, first-order logic or context-free grammar ules.sThirdly, this would bring the PAC-learning model into line with other approachessdeveloped from algorithmic omplexity theory =-=[40, 6, 2]-=- and Bayesian statistics.sAccording to Bayes' LawsPr(HIO ) = Pr(H)'Pr(OIH)sPr(O)sIf we treat bit-encoded escriptions as though they were the outcome of repeat-sedly tossing an unbiased coin, the prior... |

54 | Scientific knowledge discovery using inductive logic programming - Muggleton - 1999 |

53 | Discovering rules from large collections of examples: a case study - Quinlan - 1979 |

49 | Uncertainty, belief and probability
- Fagin, Halpern
- 1989
(Show Context)
Citation Context ...criptions of sampling distributions (eg. stochastic grammars (Lari & Young 1990) and Hidden Markov Models (HMMs)) and b) declarative representations of uncertain statements (eg. probabilistic logics (=-=Fagin & Halpern 1989-=-) and Relational Bayes’ nets (Jaeger 1997)). Stochastic Logic Programs (SLPs) (Muggleton 1996) were introduced originally a way of lifting stochastic grammars (type a representations) to the level of ... |

44 | Learning qualitative models of dynamic systems
- Coghill, Bratko, et al.
- 1992
(Show Context)
Citation Context ...The target program is specified by examples of its input/output vectors. A common ILP exercise of this kind is the induction of the quick-sort program from examples, saying for instance that the list =-=[4,1,2]-=- sorts into [1,2,4]. Suitable background knowledge contains the definition of the predicates for list concatenation, and for partitioning of a list, with respect to some value, into the lists of “smal... |

43 | The CN2 algorithm - Clark, Niblett |

38 |
Inductive Acquisition of Expert Knowledge
- Muggleton
- 1990
(Show Context)
Citation Context ... Learning have derived from systems which construct hypothesesswithin the limits of propositional logic. The major successes here have been in thesarea of inductive construction ofexpert systems (see =-=[27]-=-). The properties of varioussexpert systems are given in Figure 1. The first two, MYCIN [38] and XCON [10]swere built using hand-coding of rules. The second two, GASOIL [39] and BMT [15]swere built us... |

33 | Loglinear models for first-order probabilistic reasoning
- CUSSENS
- 1999
(Show Context)
Citation Context ...hastic Logic Programs (SLPs) (Muggleton 1996) were introduced originally a way of lifting stochastic grammars (type a representations) to the level of first-order Logic Programs (LPs). Later Cussens (=-=Cussens 1999-=-) showed that SLPs can be used to represent undirected Bayes’ nets (type representations). SLPs are presently used (Muggleton 2000) to define distributions for sampling within Inductive Logic Programm... |

32 |
Inductive learning applied to program construction and verification
- Bratko, Grobelnik
- 1992
(Show Context)
Citation Context ...tions that are true inside program loops, called loop invariants. In general, the construction of loop invariants is considered difficult, and is usually done simply by guessing. Bratko and Grobelnik =-=[3]-=- explored the idea that ILP techniques can be used for automatically constructing loop invariants. A program that is to be proved correct can be executed, and the resulting execution traces can be use... |

29 |
an oracle based approach to constructive induction
- Duce
- 1987
(Show Context)
Citation Context ...ve.s9 Strong bias ofsvocabulary. Present inductive systems construct hypothe-sses within the limits of a fixed vocabulary of propositional ttributes. Ansincreasing amount of Machine Learning research =-=[24, 28, 1, 44, 35, 16, 19]-=-sis concerned with algorithms capable of inventing auxiliary predicates whensinsufficient background knowledge is provided.s2 Inductive Logic ProgrammingsA growing body of researchers have started to ... |

25 |
randomness and incompleteness. Papers on algorithmic information theory
- Information
- 1987
(Show Context)
Citation Context ...be it propositional logic, first-order logic or context-free grammar ules.sThirdly, this would bring the PAC-learning model into line with other approachessdeveloped from algorithmic omplexity theory =-=[40, 6, 2]-=- and Bayesian statistics.sAccording to Bayes' LawsPr(HIO ) = Pr(H)'Pr(OIH)sPr(O)sIf we treat bit-encoded escriptions as though they were the outcome of repeat-sedly tossing an unbiased coin, the prior... |

24 |
Completing logic programs by inverse resolution
- Wirth
- 1989
(Show Context)
Citation Context ...ve.s9 Strong bias ofsvocabulary. Present inductive systems construct hypothe-sses within the limits of a fixed vocabulary of propositional ttributes. Ansincreasing amount of Machine Learning research =-=[24, 28, 1, 44, 35, 16, 19]-=-sis concerned with algorithms capable of inventing auxiliary predicates whensinsufficient background knowledge is provided.s2 Inductive Logic ProgrammingsA growing body of researchers have started to ... |

20 |
Application of machine learning to structural molecular biology
- Sternberg, King, et al.
- 1994
(Show Context)
Citation Context ...e understanding of the inter-relationships of chemical formula, threedimensional structure, and function of molecules of biological importance. An overview of such applications of ILP can be found in =-=[23]-=-. These ILP applications involved applying Golem [17] to protein secondary structure prediction [14], prediction of ß-sheet topology and modeling the structure activity relationship (QSAR) of a series... |

19 |
Classification of river water quality data using machine learning
- Dˇzeroski, Dehaspe, et al.
- 1994
(Show Context)
Citation Context ...t species have different sensitivity to pollutants, and therefore the structure of the macroinvertebrate community in a river is well correlated with the degree and type of pollution. Dzeroski et al. =-=[8]-=- used ILP to analyze the relation between the samples of macro-invertebrates and the quality class of water. For learning, they used 292 field samples of 68 November 1995/Vol. 38, No. 11sCOMMUNICATION... |

18 | Strategy for Constructing New Predicates in First Order Logic
- Muggleton, A
- 1988
(Show Context)
Citation Context ...ive B as in Shapiro [37] or the least general relative to B as in Plotkin [31].sAlternatively, H can be chosen to be that which produces the maximum informa-stion compression of O relative to B as in =-=[26]-=-. The choice of which constraint tosapply is necessarily tied to our notions of justification of hypotheses.s3.3 Induct ive inference and just i f icationsAs already stated, inductive inference involv... |

18 | A study of generalisation in logic programs - Niblett - 1988 |

14 |
The logic of induction
- Mortimer
- 1988
(Show Context)
Citation Context ...derpinning to the prob-slem of inductive inference. Various difficulties and paradoxes were encounteredswith these approaches which meant that they were never applied within machineslearning programs =-=[23]-=-.s3.4 PAC- learningsOne popular machine learning approach to the problem of constructing highly prob-sable hypotheses i the PAC (Probably Approximately Correct) model of learningsproposed by Valiant [... |

13 | Inductive logic programming: issues, results and the LLL challenge - Muggleton - 1999 |

11 | Lecture on the Automatic Computing Engine - Turing - 1947 |

10 |
Inverting the Resolution Principle
- Muggleton
- 1991
(Show Context)
Citation Context ...l and Puget [35] have also described related methodssfor "inventing" new predicates. However, although predicate invention has beensdemonstrated on large scale problems within a propositional setting =-=[24, 25]-=- this issnot yet the case for any first order learning systems.sRecently Quinlan [33] has described a highly efficient program, called FOIL,swhich induces first order Horn clauses. The method relies o... |

8 | Learning relations: Comparison of a symbolic and a connectionist approach (Technical Report - Quinlan - 1989 |

8 | Engineering expert system applications - Slocombe, Moore, et al. - 1986 |

7 |
A simple and general solution for inverting resolution
- Rouveirol, C, et al.
- 1989
(Show Context)
Citation Context |

7 | Loglinear models for probabilistic reasoning - Cussens - 1999 |

5 |
Learning in the limit in a growing language
- Banerji
- 1987
(Show Context)
Citation Context |

5 |
Applications of machine learning: Towards knowledge synthesis
- Bratko
- 1993
(Show Context)
Citation Context ...The target program is specified by examples of its input/output vectors. A common ILP exercise of this kind is the induction of the quick-sort program from examples, saying for instance that the list =-=[4,1,2]-=- sorts into [1,2,4]. Suitable background knowledge contains the definition of the predicates for list concatenation, and for partitioning of a list, with respect to some value, into the lists of “smal... |

4 |
Markus--An Optimized Model Inference System
- Grobelnik
- 1992
(Show Context)
Citation Context ... types at the target language level. For example, sets can be reified into lists. In [3] this refinement problem is formulated in the ILP framework. As an illustration, the general ILP program Markus =-=[9]-=- was used to implement by induction the set union operation from abstract, high-level specification.sInnovative design from first principles. Bratko [2] formulated an approach to innovative design as ... |

3 | The role of databases in knowledge-based systems
- Fox, McDermott
- 1986
(Show Context)
Citation Context ...he major successes here have been in thesarea of inductive construction ofexpert systems (see [27]). The properties of varioussexpert systems are given in Figure 1. The first two, MYCIN [38] and XCON =-=[10]-=-swere built using hand-coding of rules. The second two, GASOIL [39] and BMT [15]swere built using software derived from Quinlan's inductive decision tree buildingsalgorithm ID3 [32]. It should be note... |

3 |
Innovative design as learning from examples
- Bratko
- 1993
(Show Context)
Citation Context ...The target program is specified by examples of its input/output vectors. A common ILP exercise of this kind is the induction of the quick-sort program from examples, saying for instance that the list =-=[4,1,2]-=- sorts into [1,2,4]. Suitable background knowledge contains the definition of the predicates for list concatenation, and for partitioning of a list, with respect to some value, into the lists of “smal... |

2 |
Theory reduction with uncertainty: A reason for theoretical terms
- Ling, Dawes
- 1990
(Show Context)
Citation Context |

2 |
Finite-Element Mesh Design: An Engineering Domain for ILP Application
- Dolsak, Bratko, et al.
- 1994
(Show Context)
Citation Context ...an Dolsak) AIadvances in traditional AI regions. Because of these relational dependences, the ILP approach most naturally applies to the mesh design problem.sIn the application of ILP to this problem =-=[7]-=-, a structure to be partitioned is represented as (1) a set of edges, (2) the properties of the edges, and (3) relations among the edges. These properties and relations are represented as part of back... |

1 | Learning by Experimentation - Feng - 1990 |

1 |
of inductive learning
- Limitations
- 1989
(Show Context)
Citation Context ...e, one such constraint involves placing a constant bound k on the allowablessize of conjunctions within a boolean DNF concept description (called k-DNF). Ins304 S. Muggletonsa recent paper Dietterich =-=[7]-=- showed that for the purposes of learning DNF propo-ssitional descriptions the class of PAC-learnable concepts is highly restricted. Theresare 22" different boolean functions of m input values. Howeve... |

1 | On inductive generalisation with taxonomic background knowledge - Frisch, Page - 1989 |

1 |
News from Brainware
- Hayes-Michie
- 1990
(Show Context)
Citation Context ... systems (see [27]). The properties of varioussexpert systems are given in Figure 1. The first two, MYCIN [38] and XCON [10]swere built using hand-coding of rules. The second two, GASOIL [39] and BMT =-=[15]-=-swere built using software derived from Quinlan's inductive decision tree buildingsalgorithm ID3 [32]. It should be noted that the inductively constructed BMT is bysfar the largest expert system in fu... |

1 |
Learning simple deterministic languages. In Computational learning theory: proceedings of the second annual workshop
- Ishizaka
- 1989
(Show Context)
Citation Context |

1 |
Medicinal Chemistry 34
- Debnath, Compadre, et al.
- 1991
(Show Context)
Citation Context ...een applied to the problem of identifying Ames test mutagenicity within a series of heteroaromatic nitro compounds [15, 22]. Hansch and coworkers have studied 230 compounds using classical regression =-=[5]-=-. For 188 compounds, they successfully obtained a linear regression function using hydrophobicity, LUMO and two handcrafted binary attributes indicative of some structural properties. This regression ... |

1 |
About the Authors: IVAN BRATKO is a professor in the Faculty of Electrical Engineering and Computer
- Phys
- 1993
(Show Context)
Citation Context ...partial charge ≤ -0.022. The theory has an estimated accuracy of 89%. This matches the accuracy of both the regression analysis of Hansch and coworkers, and a more recent effort using neural networks =-=[24]-=-. It should be noted, however, that Progol’s theory is easier to comprehend and was generated automatically, without access to any structural indicator variables handcrafted by experts specifically fo... |

1 |
l)ber formal unentscheidbare Ss der Principia Mathematica und verwandter
- G5del
- 1931
(Show Context)
Citation Context ... discipline of Statistics. In turn, Statisticsswent on to have a central role in the evaluation of scientific hypotheses.sIn 1931, a year after Ghdel's previously mentioned completeness result, Gbdels=-=[12]-=- published his more famous incompleteness theorem. According to this theoremsPeano's axiomisation ofarithmetic, and any first order theory containing it, is eithersInductive Logic Programming 297sAppl... |

1 |
Explanation-based gen- eralization: A unifying view
- Mitchell, Keller, et al.
- 1986
(Show Context)
Citation Context ...reasoners make use of vast amounts of background knowledge when learning.sInductive algorithms uch as ID3 use only a fixed set of attributes attachedsto each example. Explanation-based learning (EBL) =-=[22]-=- attempted to over-scome this limitation by redefining the learning problem. In EBL hypothesessare constrained to being those derivable from background knowledge. Sincesbackground knowledge is rarely ... |

1 |
Learning relations: comparison ofa symbolic and a connectionist approach
- Quinlan
- 1989
(Show Context)
Citation Context ...ver, although predicate invention has beensdemonstrated on large scale problems within a propositional setting [24, 25] this issnot yet the case for any first order learning systems.sRecently Quinlan =-=[33]-=- has described a highly efficient program, called FOIL,swhich induces first order Horn clauses. The method relies on a general to specificsheuristic search which is guided by an information criterion ... |

1 |
Engineering expert systems applica- tions
- Slocombe, Moore, et al.
- 1986
(Show Context)
Citation Context ...tion ofexpert systems (see [27]). The properties of varioussexpert systems are given in Figure 1. The first two, MYCIN [38] and XCON [10]swere built using hand-coding of rules. The second two, GASOIL =-=[39]-=- and BMT [15]swere built using software derived from Quinlan's inductive decision tree buildingsalgorithm ID3 [32]. It should be noted that the inductively constructed BMT is bysfar the largest expert... |

1 |
The automatic omputing engine. Lecture to the London Mathe- matical Society
- Turing
- 1947
(Show Context)
Citation Context ...overy prompted Turing to attempt to show [42] that problemssconcerning incompleteness of logical theories could be overcome by the use of ansoracle capable of verifying underivable statements. Turing =-=[43]-=- later came to believesthat GSdel's incompleteness theorem required that intelligent machines be capablesof learning from examples.sVarious logical positivists including Carnap [5] have developed stat... |