|
简介：
|
A basic condition for efficient transfer learning is the similarity between the target model and relevant source models. In practice, however, the similarity condition is difficult to meet or is often violated. In this paper, under the framework of exponential family with heterogeneous source models, the related models are precisely combined by bran-new measures: linear correlation ratios between the target model and source models. Based on this type of combinations, the precision transfer likelihood is constructed by the target likelihood combined with the transferred likelihoods from the source models. Methodologically, some techniques are suggested for transferring the information from simple source models to a relatively complex target model. Theoretically, the asymptotic properties, including the standard convergence rate, are achieved, even for the case where the source models are unrelated to the target model. It can be seen from the theories and numerical results that the inference on the target model is significantly improved by the information from source models, and it is somewhat surprising that phenomenon of Stein's paradox is illustrated. |
