Sharing learned models among remote database partitions by local meta-learning
Philip K. Chan, Salvatore J. Stolfo · 1996
We explore the possibility of importing "blackbox " models learned over data sources at remote sites to improve models learned over locally available data sources. In this way, we may be able to learn more accurate knowledge from globally available data than would otherwise be possible from partial, locally available data. Proposed meta-learning strategies in our previous work are extended to integrate local and remote models. We also investigate the effect on accuracy performance when data overlap among different sites. Introduction Much of the research in inductive learning concentrates on problems with relatively small amounts of data residing at one location. With the coming age of very large network computing, it is likely that orders of magnitude more data in databases at various sites will be available for various learning problems of real world importance. Frequently, local databases represent only a partial view of all the data globally available. For example, in detecting cr...