Yazar "Karagoz, Pinar" seçeneğine göre listele

Listeleniyor 1 - 2 / 2
  • Küçük Resim Yok
    Öğe
    Policy-based memoization for ILP-based concept discovery systems
    (SPRINGER, 2016) Mutlu, Alev; Karagoz, Pinar
    Inductive Programming Logic (ILP)-based concept discovery systems aim to find patterns that describe a target relation in terms of other relations provided as background knowledge. Such systems usually work within first order logic framework, build large search spaces, and have long running times. Memoization has widely been incorporated in concept discovery systems to improve their running times. One of the problems that memoization brings to such systems is the memory overhead which may be a bottleneck. In this work we propose policies that decide what types of concept descriptors to store in memotables and for how long to keep them. The proposed policies have been implemented as extensions to a concept discovery system called Tabular CRIS wEF, and the resulting system is named Policy-based Tabular CRIS. Effects of the proposed policies are evaluated on several datasets. The experimental results show that the proposed policies greatly improve the memory consumption while preserving the benefits introduced by memoization.
  • Küçük Resim Yok
    Öğe
    Utilizing Coverage Lists as a Pruning Mechanism for Concept Discovery
    (SPRINGER-VERLAG BERLIN, 2014) Mutlu, Alev; Dogan, Abdullah; Karagoz, Pinar
    Inductive logic programming (ILP)-based concept discovery systems lack computational efficiency due to the evaluation of the large search spaces they build. One way to tackle this issue is employing pruning mechanisms. In this work, we propose a two-phase pruning mechanism for concept discovery systems that employ an Apriori-like refinement operator and evaluate the goodness of the concept descriptors based on their support value. The first step, which is novel in this work, is computationally inexpensive and prunes the search space based on the coverages of the concept descriptors. The second step employs a widely employed pruning mechanism based on the support value of the concept descriptors. The experimental results show that the first step leaves a search space reduced by 4-22% to be evaluated by the second step, which is more costly.