Context-based ensemble model

Abstract

The paper represents inference model that combines results from several different data sources. The model makes exclusive use of context properties as context is used to determine the mode different sources are combined. The mode of combining is dynamically set regarding both global and local characteristics of the context. This complicated combination schema is used in order to achieve better accuracy. The model is represented by factor graph with tree structure. That is a flexible implementation, which allows us to incorporate context usage in well-defined manner throughout inference process.