Lavrac, N., Dzeroski, S., and Grobelnik, M. (1991). Learning Non-Recursive Definitions of Relations with LINUS. In Kodrato#, Y., editor, Proceedings of the 5th European Working Session on Learning (EWSL-91), pages 265--281. Springer, Berlin.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....This may require the learning of multiple overspecific rules for covering a set of examples that could otherwise be covered by a single general one. Various systems have been developed to learn from imperfect data (for example, FOIL [Qui90a] mFOIL [Dze91] FOIL I [IKI 96] and LINUS [LDG91b] However, no system has been specially designed for learning from an incomplete background knowledge.This problem can be solved by integrating abductive reasoning into induction: abduction is used in order to complete the background knowledge by making assumptions about the incomplete ....

....an initial hypothesis. Systems of the first type are called empirical ILP systems while systems of the second type are called interactive ILP systems or incremental ILP systems [DR92] Examples of empirical ILP systems are FOIL [Qui90a] Progol [Mug95b] mFOIL [Dze91, DB92] GOLEM [MF90] LINUS [LDG91b] and TRACY [BG94b] Examples of interactive ILP systems are MIS [Sha83] MARVIN [SB86] CLINT [DRB89, DRB92b] CIGOL [MB92] and FILP [BG93] 3.2.3 Imperfect Data Real world data is often imperfect, i.e. the examples and or the background knowledge may contain various kinds of errors, either ....

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Y. Kodratoff, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence, pages 265--281. Springer-Verlag, 1991.

.... are hardwired into the system while some other can be user defined (declarative bias) Examples of hardwired restrictions are: function free programs (FOIL [Qui90a] or determinacy (GOLEM [MF90] Examples of user defined restrictions are: types and symmetry of predicates in pairs of arguments [LDG91a] input output modes [Sha83] program schemata or rule models [Wro88, Mor91] clause sets [BG95] parametrized languages [DR92] integrity constraints [DRBM91] and determinations [Rus89] In the following, we will consider only a very simple bias in the form of a set of literals which are ....

....data but this is a different problem where the methods used can not always be applied as effectively to missing information. Most of the machine learning systems that deal with incomplete information are attribute value learners. An ILP system for learning with incomplete information is LINUS [LDG91a] but, again, it essentially relies on an attribute value representation. In general, these systems adopt different methods to first complete the missing information and then learn from the completed data. In contrast, in ACL the incomplete information is handled dynamically within the learning ....

[Article contains additional citation context not shown here]

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Y. Kodratoff, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence, pages 265--281. Springer-Verlag, 1991.

....vyskyt predikatoveho symbolu nasledovaneho argumenty v zavorkach, napr. vozidlo(NEH,VOZ) Tak naprklad predstavmeli si, ze mame jeste k dispozici databazi hodnot SPZ a jim prslusejcch kraju, pak atribut Odlisny prstup k dosazen stejneho cle sleduje metoda propozicionalizace v systemu Linus [9, 10]. Zde ovsem uzivatel nema prmou kontrolu nad semantikou vyslednych atributu. Presneji receno jeho podmnoziny Datalog, ktera neuzva funkcnch symbolu krome konstant a pro nas ucel postac. Pouzit Prologu resp. Datalogu v ramci konstrukce atributu ovsem nemen nic na skutecnosti, ze vysledna ....

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Y. Kodrato#, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence, pages 265--281. Springer-Verlag, 1991.

....into a single relation to which propositional learning algorithms can be applied to was the creation of new attributes in a central relation that summarizes or aggregate information from other tables in the relational database. Variants of this method have been used in systems such as LINUS (Lavrac et al. 1991) and DINUS (Lavrac and Dzeroski, 1994) successor of the previous one) These two systems are examples of a class of ILP methods that transform restricted ILP problems into attribute value form through the technique called proposionalization and solve the transformed problem with a propositional ....

Lavrac, N., Dzeroski, S., and Grobelnik, M. Learning nonrecursive definitions of relations with LINUS. In Proceedings of the 5 th European Working Session on Learning, Springer, 1991.

....order representations and relational structures as propositions and thus supports the use of better learning algorithms, including general purpose propositional algorithms and probabilistic algorithms over the elements of the language. This approach extends previous related constructions from ILP [Lavrac et al. 1991; Khardon et al. 1999] but technically is more related to the latter. In the rest of this section we present the main constructs of the language. We omit many of the standard definitions and concentrates on the unique characteristics of R. See, e.g. Lloyd, 1987] for general details. The ....

....The basic building blocks of the representations we use (our formulae) are the same as those used by ILP representations. As presented in Sec. 2, for a predicate R and elements a; b we are not representing only the ground term R(a; b) but also R(X;Y ) R(X; b) etc. This is similar to the work in [Lavrac et al. 1991] only that our structural operations allow us to avoid some of the determinacy problems of that approach. Learning: The relational features generated by our RGFs provide a uniform domain for different learning algorithms. Applying different algorithms is easy and straightforward. Moreover, it is ....

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Machine Learning (EWSL-91), volume 482 of LNAI, pages 265-- 281, Porto, Portugal, 1991. Springer Verlag.

....extremely large and impractical to handle. The second way of transforming a relational database into a single table involves the creation of new attributes in the central fact table that summarise or aggregate information which can be found in other tables. This method is used in the LINUS system [14, 15] among others. However, it produces very wide tables with lots of data being repeated. Although more data is produced, a lot of information about how the data was originally structured is lost, and along with that the main source for efficiency in multi relational data mining. Also, the creation ....

Lavrac, N., Dzeroski, S., and Grobelnik, M. Learning nonrecursive definitions of relations with LINUS, Proceedings Fifth European Working Session on Learning, Springer, Berlin, 1991

....cannot get rid of the inherent limitations of LP regarding numerical variables: functions are not interpreted, i.e. they act as functors in terms. The consequences for that are detailed in section 2.1. Another possibility consists in mapping an ILP problem into an attribute value induction problem [8, 2, 25]. This paper investigates a radically different approach in order to handle numerical variables correctly, namely the use of Constraint Logic Programming (CLP) instead of LP as representation language. Indeed, in the past 10 years, CLP has been developed as an extension of LP to other computation ....

.... mutagenesis problem is presented in [21] 8 Discussion and Perspectives This section first discusses our choice of a maximally discriminant induction, then situates this work with respect to some previous works devoted to generalization of constraints [16, 12] or reformulation of ILP problems [8, 25]. 8.1 Generalization Choices This work first extends the frame of induction to constraint logic programming; an application, hopefully demonstrating the potentialities of this frame, is presented in [21] Note that this frame does not allow to learn clauses that could not be learned by ....

[Article contains additional citation context not shown here]

N. Lavrac, S. Dzeroski, and M. Grobelnick. Learning non recursive definitions of relations with LINUS. In Proceedings of EWSL'91, 1991.

.... case [ Quinlan, 1990 ] Many learning algorithms either build rules, or have a stage consisting in searching for rules that are postprocessed, or are aimed at producing formulae that can be easily translated into rules, Brunk and Pazzani, 1991; Cohen, 1995; Cohen, 1993; de Raedt, 1992; Lavrac et al. 1991; Muggleton and Feng, 1994; Nock and Gascuel, 1995; Pagallo and Haussler, 1990; Quinlan, 1995; Quinlan, 1994; Quinlan, 1990; Rouveirol, 1992; Thrun et al. 1991; Wnek and Michalski, 1991 ] and many others. Practical and theoretical results often focus on an aspect of approximation they both share ....

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning non-recursive definitions of relations with linus. In European Working Session in Learning, 1991.

....instead a set of first order conditions, that are used then as attributes in a classical attribute value naive Bayesian classifier. It can thus be classified as a propositionalisation approach (although the propositionalisation is done dynamically, not in a pre processing step as in LINUS [5]) In this paper we study an extension of the 1BC approach that is less propositional and directly considers probability distributions on structured individuals made of sets, tuples and multisets. We start by considering probability distributions over first and higher order terms in Section 2. ....

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Y. Kodratoff, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence, pages 265--281. SpringerVerlag, 1991.

....instead a set of first order conditions, that are used then as attributes in a classical attribute value naive Bayesian classifier. It can thus be classified as a propositionalisation approach (although the propositionalisation is done dynamically, not in a pre processing step as in LINUS [4]) In this paper we study an extension of the 1BC approach that is less propositional and directly considers probability distributions on structured individuals made of sets, tuples and multisets. We start by considering probability distributions over first and higher order terms in Section 2. ....

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Y. Kodratoff, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence, pages 265--281. SpringerVerlag, 1991.

....as they may need to correct earlier induced hypotheses. Incremental ILP systems include MIS [125] CLINT [107] MOBAL [54] FORTE [118] RX [134] LFP [144] and CIGOL [89] Nonincremental systems include GOLEM [90] FOIL [102] FOCL [95] GRENDEL [24] CLAUDIEN [113] mFOIL [32] and LINUS [66]. 10.1.2. Interactive Non interactive In interactive ILP, the learner is allowed to pose questions to an oracle (i.e. the user) about the intended interpretation. Usually these questions query the user for the intended interpretation of an example or a clause. The answers to the queries allow to ....

....it is hard to get enough examples in which the prediction is an exact number, such as 8. Instead we would like the rules to predict an interval such as mesh(Obj; X) 7 X 9; connected(Obj; Obj1) This kind of construction is not handled elegantly by existing systems (though LINUS [66] and more recently FOIL [104] can use TDIDT extensions [101] to introduce tests such as X 9) In statistics this problem of numerical prediction is known as regression. Many efficient statistical algorithms exist for handling numerical data. ILP system designers are starting to look at smoothly ....

[Article contains additional citation context not shown here]

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning non-recursive definitions of relations with LINUS. In Yves Kodratoff, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence. Springer-Verlag, 1991.

....to compare Fossil s performance to the performance of Foil. In the first series we compared the behavior of the two systems with 10 training sets of 100 instances each at different noise levels, which has been the standard procedure for evaluating many ILP systems [Quinlan, 1990, Dzeroski and Lavrac, 1991, Dzeroski and Bratko, 1992b, Muggleton et al. 1989] In the second experiment we evaluated both programs at a constant noise level of 10 , but with an increasing number of training instances. According to the results of the previous experiments we set C = 0:3 and never changed this setting. ....

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Proceedings of the European Working Session on Learning, Porto, Portugal, 1991.

....specialization (w.r.t. subsumption) of the standard model if the qualitative variable diff is provided [24] Following [4] other authors have worked on this problem. Although the U tube looks relatively simple, Bratko et al. 4] and Dzeroski [14] report how ILP systems like Foil [35] and Linus [18] are not suited to the task. Each variable has 4 landmark values and 3 time intervals, which gives 7 possible qualitative values for each variable. Combining these with the three possible directions of change give 21 possible qualitative values for each variable. With three variables, the total ....

Lavrac, N., Dzeroski, S., & Grobelnik, M. (1991). "Learning nonrecursive definitions of relations with linus", Proceedings of the European Working Session on Learning, (265--281), Berlin: Springer-Verlag.

....have to be distinguished : methods for the proper construction of the program and validation, which arises at different steps of the construction. We propose here a short survey of both aspects. 2. 1 Construction of the program Except for systems which use multiple representations such as LINUS [18], and the system TRACY [2] which learns a set of clauses at the same time, most systems in ILP only make two kinds of changes in the program : either a new clause is added, or a clause is deleted. Then differences between systems come from basic operations they use to produce a program. Usual ....

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning non-recursive definitions of relations with LINUS. In Yves Kodratoff, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence. Springer-Verlag, 1991.

....of this work was to set up the learning framework in such a way that it can be easily extended in order to face deficiencies of existing systems (e.g. working with recursive background theories, dealing with noisy data, etc. What differentiates KBI from other relational learning systems (e.g. [3, 5, 6, 12, 16, 17, 18, 19, 23]) is not a brand new search algorithm, but, rather, a more informative search structure, called LT Tree (Learned Theory Tree) that enables the system to learn a structured knowledge base limiting the complexity of the search process. The LT Tree is not a new concept: Feldman et al. 7] addressed ....

Lavrac , N., Dz eroski, S., and Grobelnik, M. "Learning non-recursive definitions of relations with LINUS", Proc. of the EWSL-91, LNAI 482, Springer-Verlag, Porto, Portugal, 1991.

....formalism however raises two major questions: that of dealing with numerical values, and that of mastering the computational complexity pertaining to first order logic. Handling numbers in ILP has mainly been tackled via transformation of relational problems into propositional ones a la LINUS [15] (see also [37] or by using adequate numerical knowledge , be it built in as in FOIL [25] or provided in declarative form as in PROGOL [20] A third possibility is based on Constraint Logic Programming (CLP) which both subsumes logic programming (LP) and allows for the interpretation of ....

N. Lavrac, S. Dzeroski, and M. Grobelnick. Learning non recursive definitions of relations with LINUS. In Proceedings of EWSL'91, 1991.

....between the attribute value and the threshold with the comparison of the difference between the attribute and the goal and the threshold. The idea of using a pre and post processing with a call to a propositional learner to do the induction step is similar to the procedure used in the LINUS system [LDG91] for learning definitions of relations. There are, however, one uninteresting feature of the two models ( Early and Goals models) Both models represent the control skills as an indexing mechanism that translates a circumstance or set of circumstances into a number (a value for a control ....

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning nonrecursive definitions of relations with linus. In Proceedings of the Fifth European Working Session on Learning, pages 265--281. Springer, Berlin, 1991.

....number handling capabilities of CLP, without requirement for additional background theory. 1 Introduction This paper is devoted to learning from positive and negative examples expressed in first order logic. Many learners have been developed in the field of Inductive Logic Programming (ILP) see [18, 22, 3, 24, 2, 13, 19], among others. However, Logic Programming (LP) and consequently ILP, does not allow for efficient handling of numbers: as emphasized by Saraswat [26] all concepts and operations of interest in [their] underlying domain of computation must be explicitly encoded in the form of Herbrand terms and ....

....for binary constraints is to enable more compact generalizations, whenever discriminant features actually involve a correlation between variables. The binary constraint based discrimination is brought back to domain constraint based discrimination, by considering auxiliary variables in the line of [13], termed relational variables. Relational variables encode the relationship between the instantiations of two variables; this enables to handle binary constraints through domain constraints defined on relational variables. Let X and Y be two variables with same initial domain of instantiation ....

[Article contains additional citation context not shown here]

N. Lavrac, S. Dzeroski, and M. Grobelnick. Learning non recursive definitions of relations with linus. In Proceedings of EWSL'91, 1991.

.... [27, 23] and Idestam Almquist s inverse implication [15] Two contending semantics of ILP have been developed [31] These are the so called open world semantics [31] and closedworld semantics [13] A number of efficient ILP systems have been developed, including FOIL [38] Golem [28] LINUS [21], CLAUDIEN [39] and Progol [26] The use of a relational logic formalism has allowed successful application of ILP systems in a number of domains in which the concepts to be learned cannot easily be described in an attribute value, propositional level, language. These applications (see Section 3) ....

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning non-recursive definitions of relations with LINUS. In Yves Kodratoff, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence. Springer-Verlag, 1991.

....theoretical 100,000 positive and negative examples, so the examples not present cannot simply be assumed to be negated. A indeterminate relational representation allows our model to be compact and homogenous (a single three placed predicate for may operate rather than 10,000 5 The LINUS system [LDG91] performs this type of transformation from relational to attribute value format. What online ML can do for KA A Case Study 12 Figure 2: The sort taxonomy with user sorts What online ML can do for KA A Case Study 13 one placed predicates) and independant of the number of objects in the ....

Nada Lavrac, Saso Dzeroski, and Marko Grobelnik. Learning nonrecursive definitions of relations with linus. In Yves Kodratoff, editor, Proc. Fifth European Working Session on Learning (EWSL), pages 265--281. Springer, 1991.

....receives a large collection of positive and negative examples from real world databases as well as background knowledge in the form of relations. The prototypical example for this research is Foil [Quinlan, 1990] and its various successors, but there are several other approaches like LINUS [Lavrac et al. 1991] and Golem [Muggleton and Feng, 1990] Theory Revision systems are not so much concerned with the induction of a useful theory, but with the maintenance of a complete and consistent theory. Theory revision systems constantly monitor the performance of their theory and attempt to generalize it ....

.... overview of the history of the field can be found in [Sammut, 1993] a selection of some of the most important papers in [Muggleton, 1992] Lavrac and Dzeroski, 1993] is an introductory book on Inductive Logic Programming with a strong focus on relational learning systems, in particular on LINUS [Lavrac et al. 1991] and mFoil [Dzeroski and Bratko, 1992a] Other introductory texts include [De Raedt and Lavrac, 1993] Muggleton, 1993] and [Muggleton and De Raedt, 1994] 2.1 Relational Learning The main concern of this thesis will be Relational Learning or Empirical Inductive Logic Programming. Learning ....

[Article contains additional citation context not shown here]

Nada Lavrac, Saso Dzeroski, and Marko Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Proceedings of the European Working Session on Learning, Porto, Portugal, 1991.

....for dynamically reducing the size of the hypothesis space. A notable exception is [Lavra c et al. 1995] where an approach for propositional literal selection [Gamberger, 1995] is used in a first order framework by transforming the first order problem into a propositional representation [Lavra c et al. 1991] . Another simple technique for first order literal selection is used in [Cohen, 1995b] where all relations are discarded which refer to objects that occur with a low frequency in the training set. However, both approaches seem to be limited to a subclass of ILP learning problems. A major ....

Nada Lavrac, Saso Dzeroski, and Marko Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Proceedings of the 5th European Working Session on Learning (EWSL-91), pages 265--281, Porto, Portugal, 1991. Springer-Verlag.

....from real world data. Significant effort has been made into investigating the effect of noisy data on attribute value learning algorithms (see e.g. 22, 3, 4, 17] Not surprisingly, noise handling methods have also entered the rapidly growing field of Inductive Logic Programming [15] Linus [16] relies directly on the noise handling abilities of decision tree learning algorithms, others, like mFoil [9] and REP [5] have adapted well known methods from attribute value learning for the ILP framework. This paper presents Fossil, a Foil like algorithm [20] that uses a search heuristic based ....

Nada Lavrac, Saso Dzeroski, and Marko Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Proceedings of the European Working Session on Learning, Porto, Portugal, 1991.

....efficiency to express concept descriptions in full first order logic. Figure 3 summarizes the multiple representation approach used in LINUS, which is a framework for relational learning algorithms that supports efficient learning behavior by solving learning tasks with monadic learning algorithms (Lavrac, Dzeroski, Grobelnik, 1991). It inputs instances in the Deductive Hierarchical Database (DHDB) representation, which can express typed nonrecursive Horn clauses with negation (Mozetic, 1987) Whereas Wyl reduces the relational learning problem into one where explanation based learning techniques can be applied, LINUS ....

Lavrac, N., Dzeroski, S., & Grobelnik, M. (1991). Learning nonrecursive definitions of relations with LINUS. In Proceedings of the Fifth European Working Session on Learning (pp. 265--281). Porto, Portugal: Springer-Verlag.

No context found.

N. Lavrac, S. Dzeroski and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Proc. Fifth European Working Session on Learning, pp. 265--281. Springer.

No context found.

Lavrac, N. Dzeroski, S. and Grobelnik, M., "Learning nonrecursive definitions of relations with LINUS." In: Proceedings of the 5th European Working Session on Learning. Springer, 1991, pp. 265--281.

....3.2) As input it takes training examples E , given as ground facts, and background knowledge T in the form of (possibly recursive) deductive database (DDB) clauses. The main idea of LINUS is to transform the problem of learning relational DHDB descriptions into an attribute value learning task [27]. This is achieved by the so called DHDB interface. The interface transforms the training examples from the DHDB form into the form of attributevalue tuples. This results in an extensional table, as introduced in Section 2.2.2. The most important feature of this interface is that, by taking into ....

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Y. Kodratoff, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence, pages 265--281. Springer-Verlag, 1991.

.... may define the learning setting so that the task of finding a hypothesis depends on E and the literal space L, such that H = H(E;L) 3 Preprocessing Preprocessing consists of two steps: First, the transformation of the relational learning task into an attribute value learning task by the LINUS [9, 11] transformer: by taking into account the types of the arguments of the target predicate, the applications of utility predicates and functions are considered as attributes for learning by an attribute value learner. This step is illustrated below. The second step is the generation of negative ....

....by two relations, adjacent(X; Y ) and less than(X; Y ) indicating that rank file X is adjacent to rank file Y and rank file X is less than rank file Y , respectively. The arguments of these background predicates are not typed and the same predicates are used for both types of arguments. In LINUS [9, 11], each of these predicates is replaced by two predicates, one for each type of arguments. Thus, LINUS uses the following relations: adjacent f ile(X; Y ) and less f ile(X; Y ) with arguments of type f ile (with values a to h) adjacent rank(X; Y ) and less rank(X; Y ) with arguments of type rank ....

[Article contains additional citation context not shown here]

N. Lavrac, S. Dzeroski and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Proc. Fifth European Working Session on Learning, pages 265--281, Springer, Berlin, 1991.

....in learning The available patient data may be augmented with additional diagnostic knowledge which can be considered as additional information for the learner. In machine learning terminology, additional expert knowledge is usually referred to as background knowledge. The main idea in LINUS [32, 31] is to incorporate different attribute value learning algorithms into an environment for inductive logic programming [40, 31] which enables the effective use of specialist background knowledge in learning. LINUS also enables the induction of relational descriptions. Several attribute value ....

Lavrac, N. Dzeroski, S., Grobelnik, M. (1991) Learning nonrecursive definitions of relations with LINUS. In Proceedings of the 5th European Working Session on Learning. Springer, pp. 265--281.

....and Bratko 1992] Some of this work relies on inductive logic programming [Lavrac and Dzeroski 1993] systems to induce the models from example behaviors and background knowledge consisting of the definitions of the QSIM [Kuipers 1986] constraints. The LINUS approach to inductive logic programming [Lavrac et al. 1991] is based on the idea of transforming an inductive logic programming problem to propositional form and then applying propositional learning systems. This is accomplished by using background knowledge predicates to introduce new variables and generate propositional features [Dzeroski et al. 1992] ....

Lavrac, N., Dzeroski, S., and Grobelnik, M. (1991). Learning nonrecursive definitions of relations with LINUS. In Proc. Fifth European Working Session on Learning, pages 265--281. Springer, Berlin.

No context found.

Lavrac, N., Dzeroski, S., and Grobelnik, M. (1991). Learning nonrecursive definitions of relations with LINUS. In Proc. Fifth European Working Session on Learning, pages 265-- 281. Springer, Berlin.

.... Inductive logic programming (ILP) refers to first order learning of relational descriptions in the representation formalism of logic programs [7] In this setting, a LINUS transformation approach is assumed that is appropriate for a limited hypothesis language of constrained nonrecursive clauses [6,7]. For example, if the training examples about the target relation daughter(A 1 ; A 2 ) are given in the training set, and the background knowledge consists of the definitions of a unary relation female and binary relation parent, the transformation of training examples results in a matrix of ....

....for a selected feature, one of the two complementary literals (positive or negative) is used to construct a rule. We suggest that learning should be based on tuples of truth values of positive and negative literals, rather than on feature vectors. RL ICET is similar to the LINUS learning system [6,7] since it uses a three part learning strategy. First, a preprocessor translates the Prolog relations and predicates into a feature vector format. The preprocessor in RL ICET was designed specially for the East West Challenge, whereas LINUS has a general purpose preprocessor. Second, an ....

Lavrac, N., S. Dzeroski and M. Grobelnik:. Learning Nonrecursive Definitions of Relations with LINUS, in: Proceedings of the 5th European Working Session on Learning, Springer, 1991, 265--281.

No context found.

Lavrac, N., Dzeroski, S. and Grobelnik, M. (1991) Learning nonrecursive definitions of relations with LINUS. Proc. of EWSL 91, pp. 265-281. Porto, Portugal, Springer-Verlag.

....program synthesis and program analysis, and vice versa. In this paper, we investigate the different faces of ILP. We first discuss two different semantics for ILP: a normal semantics for ILP introduced by Plotkin [38] and followed by Muggleton [28] and incorporated in many well known systems [29, 34, 40, 8, 9, 24], and a non monotonic semantics derived from Helft s work [19] and used in [33, 12] It is shown that the normal semantics leads to some problems when learning multiple predicates, and that these problems can be avoided using the non monotonic semantics. The latter also allows for the induction of ....

....theory B and hypotheses H are represented by sets of clauses. For simplicity, we mainly focus on definite clauses for representing hypotheses and theories. Nevertheless, parts of our discussion extend to the more general (normal) program clauses [25] which are sometimes used in ILP systems [24, 40], as well as to (general) clauses, as used in [12] Language bias imposes certain syntactic restrictions on the form of clauses allowed in hypotheses; for example, one might consider only constrained clauses [24] these are clauses for which all variables occurring in the body also occur in the ....

[Article contains additional citation context not shown here]

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning non-recursive definitions of relations with LINUS. In Yves Kodratoff, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence. Springer-Verlag, 1991.

....(RLGG) of a set of observations with respect to some background knowledge; by making certain syntactic restrictions, it can be shown that a unique, finite RLGG can be found, which is then simplified to generate a reasonable hypothesis. The third class of systems, which includes linus [Lavrac et al. 1991] and foil [Quinlan 1990] work by extending propositional approaches to a first order framework. While foil uses heuristic search techniques from propositional learning directly, linus explicitly converts the first order representation to a propositional problem by defining appropriate sets of new ....

.... specifying which arguments of a predicate are to be considered as input (old variables) and which as output (either old or new variables) is used in both foil [Quinlan 1990] and golem [Muggleton and Feng 1990] The arguments of the background knowledge predicates may also be sorted, as in linus [Lavrac et al. 1991], in which case the information about the sorts of variables (arguments) greatly reduces the number of propositional features involved. Example: For our simple ILP example, we have j = 2, i = 1, l = 2, and n = 2. A literal father(X; A) where X is old and A is new is not determinate, as a man ....

N. Lavrac, S. Dzeroski and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Proc. Fifth European Working Session on Learning, pages 265--281, Springer, Berlin, 1991.

....A new area in machine learning, called inductive logic programming [36] is concerned with the development of programs which induce relational descriptions in some restricted first order formalism. A restricted form of logic programs is used in MIS [46] CIGOL [37] GOLEM [38] FOIL [45] and LINUS [26]. So far, to our knowledge, LINUS was the only inductive logic programming system that addressed the problem of inducing medical diagnostic rules, in particular the problem of learning rules for early diagnosis of rheumatic diseases [27] It was possible to apply LINUS to these problems for two ....

Lavrac, N., Dzeroski, S. and Grobelnik, M. (1991) Learning nonrecursive definitions of relations with LINUS. Fifth European Working Session on Learning, EWSL 91. Porto, Portugal: Springer-Verlag.

No context found.

Lavrac, N., Dzeroski, S., and Grobelnik, M. (1991). Learning Non-Recursive Definitions of Relations with LINUS. In Kodrato#, Y., editor, Proceedings of the 5th European Working Session on Learning (EWSL-91), pages 265--281. Springer, Berlin.

No context found.

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Y. Kodratoff, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence, pages 265--281. SpringerVerlag, 1991.

No context found.

N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Y. Kodratoff, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence, pages 265--281. SpringerVerlag, 1991.

No context found.

Lavrac, N., Dzeroski, S., and Grobelnik, M.: Learning nonrecursive definitions of relations with LINUS. In Proceedings of the 5th European Working Session on Learning, Springer (1991)

No context found.

) N. Lavrac, S. Dzeroski, and M. Grobelnik. Learning nonrecursive definitions of relations with LINUS. In Y. Kodratoff, editor, Proceedings of the 5th European Working Session on Learning, volume 482 of Lecture Notes in Artificial Intelligence, pages 265--281. Springer-Verlag, 1991.

No context found.

LavraF N. , Dzeroski, S., & Grobelnik M., "Learning non-recursive definitions of Relations with LINUS.", in Proceedings of the European Working Session on Learning, pp. 265-281, Springer-Verlag (482), 1991.

No context found.

Lavrac N. , Dzeroski, S., & Grobelnik M., "Learning nonrecursive definitions of Relations with LINUS.", in Proceedings of the European Working Session on Learning, pp. 265-281, Springer-Verlag (482), 1991.

No context found.

Lavrac , N., Dz eroski, S., and Grobelnik, M. "Learning non-recursive definitions of relations with LINUS", Proc. of the EWSL-91, LNAI 482, Springer-Verlag, Porto, Portugal, 1991.

No context found.

Lavrac, N., Dzeroski, S., & Grobelnik, M. (1991). Learning Non-Recursive Definitions of Relations with LINUS. In Y. Kodratoff (Eds.), European Working Session on Learning. (pp. 265-281). Porto, Portugal: Springer-Verlag.

No context found.

Lavrac, N. and Dzeroski, S. and Grobelnik, M. (1991). Learning Nonrecursive Definitions of Relations with LINUS. In Y. Kodratoff, editor, EWSL-91: Proceedings of the European Working Session on Learning, pages 265--281. Springer-Verlag, Berlin.

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