ABSTRACT

The problem of distributed data mining is very important in network problems. Ina distributed environment (such as a sensor or IP network), one has distributed probes placed at strategic locations within the network. The problem here is to be able to correlate the data seen at the various probes, and discover patterns in the global data seen at all the different probes. There could be different models of distributed data mining here, but one could involve a NOC that collects data from the distributed sites, and another in which all sites are treated equally. The goal here obviously would be to minimize the amount of data shipped between the various sites — essentially, to reduce the communication overhead. In distributed mining, one problem is how to mine across multiple heterogeneous data sources: multi-database and multi-relational mining. Another important new area is adversary data mining. In a growing number of domains — email spam, counter-terrorism, intrusion detection/computer security, click spam, search engine spam, surveillance, fraud detection, shop bots, file sharing, etc. — data mining systems face adversaries that deliberately manipulate the data to sabotage them (e.g. make them produce false negatives). In this paper need to develop systems that explicitly take this into account, by combining data mining with game theory.