Boosting feature primitives for minimally supervised knowledge discovery and improved classification
Frederick E. Petry, Raja Tanveer Iqbal · 2006
The task of learning is often made difficult by adversaries specific to different domains. In general, the term 'adversary' may be used for any natural or man made agent that causes a learning algorithm to deviate from its ideal behavior. We propose learning algorithms that deal with adversaries present in different domains of application. The common characteristic of these learning algorithms is that they use feature primitives and weak learners for learning. We argue that a large number of suitably chosen feature primitives can be a very powerful tool for robust learning.