Knowledge Discovery in Adversarial Settings

摘要: In adversarial settings, the interests of modellers and those being modelled do not run together. This includes domains such as law enforcement counterterrorism but, increasingly, also more mainstream customer relationship management. The conventional strategy, maximizing fit a model to available data, does work in settings because data cannot all be trusted, it makes results too predictable adversaries. Some existing techniques remain applicable, others can used if they are modified allow for setting, while must discarded. General principles this domain discussed implications practice outlined.