Systems and techniques for distributed and stream data mining
Domenico Talia, Paolo Trunfio, Salvatore Orlando, Raffaele Perego, Claudio Silvestri · 2006
Nowadays huge amounts of electronic data are naturally collected in distributed sites, due to either plural ownership or geographical distribution of the processes that produce data. Moving data to a location for extracting useful and actionable knowledge is usually considered unfeasible, for either policy or technical reasons. It thus becomes mandatory to mine them by exploiting the multiple distributed resources close to data repositories. This requires to develop novel distributed data mining (DM) algorithms and systems, able to return a global knowledge by aggregating multiple local results. Besides the distributed nature of data under analysis, nowadays researchers have also to consider the streaming nature of them, which entails approximate on-line methods to mine data stream. This report introduces requirements, design issues, and typical solutions of distributed DM algorithms. We survey some of the most important works that recently appeared in the literature on this topic, included the latest proposals regarding P2P, highly distributed, approximate algorithms. Moreover, we discuss some of the most important proposals concerning stream DM algorithms. Such DM algorithms can be considered as the building blocks for realizing seamless distributed knowledge extraction processes and systems. Such systems can take advantage of the recent advances in computational and data Grid, which many researchers consider as the enabling technology for also developing highperformance knowledge discovery processes and systems. In this report we analyze some significative examples of distributed and Grid-oriented knowledge discovery systems.