Makoto Takaki

A Extraction Method of Overlapping Cluster Based on Network Structure Analysis

In this paper, we propose a method of overlapping clustering based on network structure analysis that improves the crisp clustering algorithm proposed by Newman et al. In the proposed technique, we cluster the nodes using Newmans algorithm. We then make a contraction graph in which a cluster is considered as a node. In addition, we cluster the created contraction graph using Newmans clustering algorithm again and identify the overlapping nodes. Overlapping clustering is more flexible than crisp clustering. The experimental results using the real network data and trackback data represented the efficacy of the proposed technique.

New Dynamic Load Balancing for Parallel Modified PrefixSpan

A motif that is a featured pattern is discovered from the frequent patterns in amino acid sequences. To extract frequent patterns at high speed, a parallel Modified PrefixSpan with a master-worker paradigm was proposed. However, a master-worker paradigm has a performance limitation when the number of PCs increases. To address this disadvantage, the distributed worker paradigm is adapted to the parallel Modified PrefixSpan. In order to obtain an effective speed-up ratio, we propose a new dynamic load balancing. The characteristics of dynamic load balancing are a smallgrain task and a Cache-based Random Steal schema. When a 100-scale PC cluster was used, the experimental results showed a speed-up ratio of 95 times.

A new dynamic load balancing technique for parallel modified PrefixSpan with distributed worker paradigm and its performance evaluation

In order to extract the frequent patterns that can become motif at high speed from amino acid sequences, we are developing the parallel Modified PrefixSpan with the distributed worker paradigm. This paper presents a new dynamic load balancing technique for the parallel Modified PrefixSpan with the distributed worker paradigm and its performance evaluation. The characteristics of the dynamic load balancing are the small-grain task and the Cache-based Random Steal schema. This paper explains these characteristics and presents performance evaluations with the PC cluster of 100 nodes.