Abstract
In this work we investigate how different machine settings for a hardware Distributed Shared Memory (DSM) architecture affect the performance of parallel logic programming (PLP) systems. We use execution-driven simulation of a DASH-like multiprocessor to study the impact of the cache block size, the cache size, the network bandwidth, the write buffer size, and the coherence protocol on the performance of Andorra-I, a PLP system capable of exploiting implicit parallelism in Prolog programs. Among several other observations, we find that PLP systems favour small cache blocks regardless of the coherence protocol, while they favour large cache sizes only in the case of invalidate-based coherence. We conclude that the cache block size, the cache size, the network bandwidth, and the coherence protocol have a significant impact on the performance, while the size of the write buffer is somewhat irrelevant.

Abstract
In this paper we use execution-driven simulation of a scalable multiprocessor to evaluate the performance of the Andorra-I parallel logic programming system under invalidate and update-based protocols. We use two versions of Andorra-I. One of them was originally designed for bus-based multiprocessors, while the other is optimised for scalable architectures. We study a well-known invalidate protocol and two different update-based protocols. Our results show that for our sample logic programs the update-based protocols outperform their invalidate-based counterpart for the original version of Andorra-I. In contrast, the optimised version of Andorra-I benefits the most from the invalidate-based protocol, but a hybrid update-based protocol performs as well as the invalidate protocol in most cases. We conclude that parallel logic programming systems can consistently benefit from hybrid update-based protocols. © Springer-Verlag Berlin Heidelberg 2000.

Abstract
Large-scale applications that require executing very large numbers of tasks are only feasible through parallelism. In this work we present a system that automatically handles large numbers of experiments and data in the context of machine learning. Our system controls all experiments, including re-submission of failed jobs and relies on available resource managers to spawn jobs through pools of machines. Our results show that we can manage a very large number of experiments, using a reasonable amount of idle CPU cycles, with very little user intervention.

Abstract
Ensembles have proven useful for a variety of applications, with a variety of machine learning approaches. While Quinlan has applied boosting to FOIL, the widely-used approach of bagging has never been employed in ILP. Bagging has the advantage over boosting that the different members of the ensemble can be learned and used in parallel. This advantage is especially important for ILP where run-times often are high. We evaluate bagging on three different application domains using the complete-search ILP system, Aleph. We contrast bagging with an approach where we take advantage of the non-determinism in ILP search, by simply allowing Aleph to run multiple times, each time choosing "seed" examples at random.

Abstract
Grid environments are ideal for executing applications that require a huge amount of computational work, both due to the big number of tasks to execute and to the large amount of data to be analysed. Unfortunately, current tools may require that users deal themselves with corrupted outputs or early termination of tasks. This becomes incovenient as the number of parallel runs grows to easily exceed the thousands. ReCS is a user-level software designed to provide automatic detection and restart of corrupted or early terminated tasks. ReGS uses a web interface to allow the setup and control of grid execution, and provides automatic input data setup. ReGS allows the automatic detection of job dependencies, through the GRID-ADL task management language. Our results show that besides automatically and effectively managing a huge number of tasks in grid environments, ReGS is also a good monitoring tool to spot grid nodes pitfalls. © 2005 IEEE.

Abstract
Learning from multi-relational domains has gained increasing attention over the past few years. Inductive logic programming (ILP) systems, which often rely on hill-climbing heuristics in learning first-order concepts, have been a dominating force in the area of multi-relational concept learning. However, hill-climbing heuristics are susceptible to local maxima and plateaus. In this paper, we show how we can exploit the links between objects in multi-relational data to help a first-order rule learning system direct the search by explicitly traversing these links to find paths between variables of interest. Our contributions are twofold: (i) we extend the pathfinding algorithm by Richards and Mooney [12] to make use of mode declarations, which specify the mode of call (input or output) for predicate variables, and (ii) we apply our extended path finding algorithm to saturated bottom clauses, which anchor one end of the search space, allowing us to make use of background knowledge used to build the saturated clause to further direct search. Experimental results on a medium-sized dataset show that path finding allows one to consider interesting clauses that would not easily be found by Aleph. © Springer-Verlag Berlin Heidelberg 2005.