ERCIM “ Alain Bensoussan ” Fellowship Scientific Report

The Affinity Propagation (AP) clustering algorithm proposed by Frey and Dueck (2007) provides an understandable, nearly optimal summary of a data set, through message passing among all pairs of data items. However, it suffers two major shortcomings: i) the number of clusters is vague with the user-defined parameter called self-confidence, and ii) the quadratic computational complexity. When aiming at a given number of clusters due to prior knowledge (e.g., number of classes in a supervised context), AP has to be launched many times until an appropriate setting of self-confidence parameter is found by a bisection method. The re-launched AP increases the computational cost by 1 order of magnitude. In this paper, we propose an algorithm, called K-AP, to exploit the immediate results of K clusters by introducing a constraint in the process of message passing. Through theoretical analysis and experimental validation, K-AP was shown to be able to directly generate K clusters as userdefined, with a negligible increase of computational cost compared to AP. In the meanwhile, KAP preserves the clustering quality as AP in terms of the distortion (sum of the squared distance between each data item and its assigned cluster center). It is more effective than k-mediods w.r.t. the distortion minimization and higher clustering purity. Adaptively detecting changes in autonomic grid computing. Xiangliang Zhang, Cecile Germain and Michele Sebag. In: Proceedings of 11th ACM/IEEE International Conference on Grid Computing (Grid 2010), workshop on Autonomic Computational Science. Abstract: Detecting the changes is the common issue in many application fields due to the non-stationary distribution of the applicative data, e.g., sensor network signals, web logs and grid running logs. Toward Autonomic Grid Computing, adaptively detecting the changes in a grid system can help to alarm the anomalies, clean the noises, and report the new patterns. In this paper, we proposed an approach of self-adaptive change detection based on the PageHinkley statistic test. It handles the non-stationary distribution without the assumption of data distribution and the empirical setting of parameters. We validate the approach on the EGEE streaming jobs, and report its better performance on achieving higher accuracy comparing to the other change detection methods. Meanwhile this change detection process could help to discover the device fault which was not claimed in the system logs. Detecting the changes is the common issue in many application fields due to the non-stationary distribution of the applicative data, e.g., sensor network signals, web logs and grid running logs. Toward Autonomic Grid Computing, adaptively detecting the changes in a grid system can help to alarm the anomalies, clean the noises, and report the new patterns. In this paper, we proposed an approach of self-adaptive change detection based on the PageHinkley statistic test. It handles the non-stationary distribution without the assumption of data distribution and the empirical setting of parameters. We validate the approach on the EGEE streaming jobs, and report its better performance on achieving higher accuracy comparing to the other change detection methods. Meanwhile this change detection process could help to discover the device fault which was not claimed in the system logs. Self-adaptive Change Detection in Streaming Data with Non-stationary Distribution. Xiangliang Zhang and Wei Wang. Accepted by International Conference on Advanced Data Mining and Applications (ADMA2010). Abstract: Non-stationary distribution, in which the data distribution evolves over time, is a common issue in many application fields, e.g., intrusion detection and grid computing. Detecting the changes in massive streaming data with non-stationary distribution helps to alarm the anomalies, clean the noises, and report the new patterns. In this paper, we employ a novel approach for detecting changes in streaming data with the purpose of improving the quality of modeling the data streams. The self-adaptability of the novel approach enhances the effectiveness of modeling data streams by timely catching the changes of distributions. We Non-stationary distribution, in which the data distribution evolves over time, is a common issue in many application fields, e.g., intrusion detection and grid computing. Detecting the changes in massive streaming data with non-stationary distribution helps to alarm the anomalies, clean the noises, and report the new patterns. In this paper, we employ a novel approach for detecting changes in streaming data with the purpose of improving the quality of modeling the data streams. The self-adaptability of the novel approach enhances the effectiveness of modeling data streams by timely catching the changes of distributions. We Page 3 sur 3 validated the approach on an online clustering framework with a benchmark KDDcup 1999 intrusion detection data set as well as a real-world grid data set. The validation results demonstrate its better performance on achieving higher accuracy and lower percentage of outliers comparing to the other change detection approaches. III -Attended Seminars, Workshops, and Conferences Please identify the name(s), date(s) and place(s) of the events in which you participated during your fellowship period.