Discriminative Structure Learning of Markov Logic Networks

Biba, Marenglen; Ferilli, Stefano; Esposito, Floriana

doi:10.1007/978-3-540-85928-4_9

Cited by 24 publications

(43 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is unclear whether it is feasible to alter LSM to efficiently learn clauses with constants since such constants may need to be considered individually which dramatically increases the search space. This problem also holds for other existing MLN structure learners [13,21,1,14].…”

Section: Resultsmentioning

confidence: 97%

See 1 more Smart Citation

Online Structure Learning for Markov Logic Networks

Huynh

Mooney

2011

Machine Learning and Knowledge Discovery in Databases

View full text Add to dashboard Cite

Abstract. Most existing learning methods for Markov Logic Networks (MLNs) use batch training, which becomes computationally expensive and eventually infeasible for large datasets with thousands of training examples which may not even all fit in main memory. To address this issue, previous work has used online learning to train MLNs. However, they all assume that the model's structure (set of logical clauses) is given, and only learn the model's parameters. However, the input structure is usually incomplete, so it should also be updated. In this work, we present OSL-the first algorithm that performs both online structure and parameter learning for MLNs. Experimental results on two realworld datasets for natural-language field segmentation show that OSL outperforms systems that cannot revise structure.

show abstract

Section: Resultsmentioning

confidence: 97%

“…logical clauses) of an MLN [13,21,1,14,15] are batch algorithms that are effectively designed for training data with relatively few mega-examples [20]. A mega-example is a large set of connected facts, and mega-examples are disconnected and independent from each other.…”

Section: Introductionmentioning

confidence: 99%

Online Structure Learning for Markov Logic Networks

Huynh

Mooney

2011

Machine Learning and Knowledge Discovery in Databases

View full text Add to dashboard Cite

show abstract

“…The precision-recall curve is computed by varying the threshold above which a ground atom is predicted to be true. Parameters for ILS-DSL,BUSL and ILS were respectively set as in [3], [13] and [1]. To guarantee the fairness of comparison, we set the maximum number of literals per clause to 5 for all systems as it is shown in [3].…”

Section: Systems and Methodologymentioning

confidence: 99%

“…This can lead to suboptimal results when these clauses do not capture the essential dependencies in the domain in order to improve classification accuracy [3]. To the best of our knowledge, there only exists two systems, that learn the structure of MLNs for a discriminative task.…”

Section: Discriminative Learning Of Mlnsmentioning

confidence: 99%

“…One first uses ALEPH [21] to learn a large set of potential clauses, then learns the weights and prunes useless clauses guided by weights [7]. The second method, called Iterated Local Search -Discriminative Structure Learning (ILS-DSL), chooses the structure by maximizing CLL and sets the parameters by maximizing WPLL [3]. Iterated Local Search is used for searching candidate clauses and for every candidate structure, the quasi-Newton optimization method L-BFGS is used to set weights optimizing WPLL.…”

Section: Discriminative Learning Of Mlnsmentioning

confidence: 99%

See 1 more Smart Citation

Discriminative Markov Logic Network Structure Learning Based on Propositionalization and χ 2-Test

Dinh

Exbrayat

Vrain

2010

Advanced Data Mining and Applications

View full text Add to dashboard Cite

Abstract. In this paper we present a bottom-up discriminative algorithm to automatically learn Markov Logic Network structures. Our approach relies on a new propositionalization method that transforms the learning dataset into an approximative representation in the form of a boolean table. Using this table, the algorithm constructs a set of candidate clauses according to a χ 2 independence test. To compute and choose clauses, we successively use two different optimization criteria, namely log-likelihood (LL) and conditional log-likelihood (CLL), in order to combine the efficiency of LL optimization algorithms together with the accuracy of CLL ones. First experiments show that our approach outperforms existing discriminative MLN structure learning algorithms.

show abstract