Knowledge transfer

摘要: People learn from prior experiences. We first how to use a spoon and then know different size of spoon. sew embroider. Transferring knowledge one situation another related often increases the speed learning. This observation is relevant human learning, as well machine This thesis focuses on problem transfer — an area study in The goal train system recognize apply acquired previous tasks new or domains. An effective facilitates learning processes for novel tasks, where little information available. For example, ability model that identifies writers born U.S. identify Kiribati, much lesser known country, would increase Kiribati scratch. In this thesis, we investigate three dimensions transfer: what, how, why. present elaborate these questions: What type transfer? How across entities? Why certain pattern observed? propose Segmented Transfer most informative partitions tasks. proposed applied Wikipedia vandalism detection entity search retrieval improves predictions. Based foundation network theory, Knowledge Network (KTN), describing relationships among problems. KTN not only representation, but also framework select efficient ensemble learners improve predictive model. provides insights identifying ontological connections were initially obscured. may observe occurs dissimilar such transferring using knife fork chopsticks.