Learning One More Thing

摘要

Most research on machine learning has focused on scenarios in which a learnerfaces a single, isolated learning risk. The life-long learning framework assumes instead that the learner encounters a multitude of related learning tasks over its lifetime, providing the opportunity for the transfer of knowledge. This paper studies life-long learning in the context of binary classification. It presents the invariance approach, in which knowledge is transferred via a learned model of the invariance of the domain. Results on learning to recognize objects from color images demonstrate superior generalization capabilities if invariances are learned and used to bias subsequent learning.