Results 1 
7 of
7
Feature Discovery with Type Extension Trees
"... Abstract. We are interested in learning complex combinatorial features from relational data. We rely on an expressive and general representation language whose semantics allows us to express many features that have been used in different statistical relational learning settings. To avoid expensive e ..."
Abstract

Cited by 5 (5 self)
 Add to MetaCart
(Show Context)
Abstract. We are interested in learning complex combinatorial features from relational data. We rely on an expressive and general representation language whose semantics allows us to express many features that have been used in different statistical relational learning settings. To avoid expensive exhaustive search over the space of relational features, we introduce a heuristic search algorithm guided by a generalized relational notion of information gain and a discriminant function. The algorithm succesfully finds interesting and interpretable features on artificial and realworld relational learning problems. 1
Relational information gain
 Machine Learning
, 2009
"... Abstract. We introduce relational information gain, a renement scoring function measuring the informativeness of newly introduced variables. The gain can be interpreted as a conditional entropy in a welldened sense and can be efciently approximately computed. In conjunction with simple greedy gen ..."
Abstract

Cited by 3 (1 self)
 Add to MetaCart
(Show Context)
Abstract. We introduce relational information gain, a renement scoring function measuring the informativeness of newly introduced variables. The gain can be interpreted as a conditional entropy in a welldened sense and can be efciently approximately computed. In conjunction with simple greedy generaltospecic search algorithms such as FOIL, it yields an efcient and competitive algorithm in terms of predictive accuracy and compactness of the learned theory. 1
An Efficient Approximation to Lookahead in Relational Learners
"... Abstract. Greedy machine learning algorithms suffer from shortsightedness, potentially returning suboptimal models due to limited exploration of the search space. Greedy search misses useful refinements that yield a significant gain only in conjunction with other conditions. Relational learners, suc ..."
Abstract

Cited by 2 (0 self)
 Add to MetaCart
(Show Context)
Abstract. Greedy machine learning algorithms suffer from shortsightedness, potentially returning suboptimal models due to limited exploration of the search space. Greedy search misses useful refinements that yield a significant gain only in conjunction with other conditions. Relational learners, such as inductive logic programming algorithms, are especially susceptible to this problem. Lookahead helps greedy search overcome myopia; unfortunately it causes an exponential increase in execution time. Furthermore, it may lead to overfitting. We propose a heuristic for greedy relational learning algorithms that can be seen as an efficient, limited form of lookahead. Our experimental evaluation shows that the proposed heuristic yields models that are as accurate as models generated using lookahead. It is also considerably faster than lookahead. 1
Mach Learn (2011) 83: 219–239 DOI 10.1007/s1099401051947 Relational information gain
"... Abstract We introduce relational information gain, a refinement scoring function measuring the informativeness of newly introduced variables. The gain can be interpreted as a conditional entropy in a welldefined sense and can be efficiently approximately computed. In conjunction with simple greed ..."
Abstract
 Add to MetaCart
(Show Context)
Abstract We introduce relational information gain, a refinement scoring function measuring the informativeness of newly introduced variables. The gain can be interpreted as a conditional entropy in a welldefined sense and can be efficiently approximately computed. In conjunction with simple greedy generaltospecific search algorithms such as FOIL, it yields an efficient and competitive algorithm in terms of predictive accuracy and compactness of the learned theory. In conjunction with the decision tree learner TILDE, it offers a beneficial alternative to lookahead, achieving similar performance while significantly reducing the number of evaluated literals.
Supervisor
"... Most of the data analysis approaches typically assume the input data stored in a single table; they cannot analyse relational data without first transforming it into a single table. This transformation, however, is not always easy and results in the lost of the structural information that could pote ..."
Abstract
 Add to MetaCart
Most of the data analysis approaches typically assume the input data stored in a single table; they cannot analyse relational data without first transforming it into a single table. This transformation, however, is not always easy and results in the lost of the structural information that could potentially be useful for the data mining processes. Relational data mining research aims to develop data analysis solutions for relational data without requiring it to be transformed into a single table. This thesis presents an investigation into relational data mining for building rulebased classifiers to predict classes of previously unseen objects which are stored in and managed by a relational database management system. In this study, the ila2 propositional rule induction algorithm was adapted to the relational domain for learning classifiers from relational data. Using the new relational learning algorithm, the rila relational learning system was developed with two rule selection strategies; the select early, and the select late strategies. In the select early strategy, inherited from the ila2 algorithm, rules are selected
Learning Type Extension Trees for Metal Bonding State Prediction
"... Abstract. Type Extension Trees (TET) have been recently introduced as an expressive representation language allowing to encode complex combinatorial features of relational entities. They can be efficiently learned with a greedy search strategy driven by a generalized relational information gain and ..."
Abstract
 Add to MetaCart
Abstract. Type Extension Trees (TET) have been recently introduced as an expressive representation language allowing to encode complex combinatorial features of relational entities. They can be efficiently learned with a greedy search strategy driven by a generalized relational information gain and a discriminant function. In predicting the metal bonding state of proteins, TET achieve significant improvements over manually curated motifs, and the expressiveness of combinatorial features significantly contributes to such performance. Preliminary collective classification results seem to indicate it as a promising direction for further research. 1 Learning Type Extension Trees A TET [1] consists of a treestructured logic formula where nodes are conjunctions of literals, and edges are labeled with sets of variables. Instead of a simple truth assignment, a TET defines a complex combinatorial feature whose recursive value structure accounts for the number of times each subtree can be satisfied, given the possible bindings of its edge variables. A simple discriminant function [2] can be defined over TET