This work studies the problem of transfer learning under the functional linear model framework, which aims to improve the fit of the target model by leveraging the knowledge from related source models. We measure the relatedness between target and source models using Reproducing Kernel Hilbert Spaces, allowing the type of knowledge being transferred to be interpreted by the structure of the spaces. Two algorithms are proposed: one transfers knowledge when the index of transferable sources is known, while the other one utilizes aggregation to achieve knowledge transfer without prior information about the sources. Furthermore, we establish the optimal convergence rates for excess risk, making the statistical gain via transfer learning mathematically provable. The effectiveness of the proposed algorithms is demonstrated on synthetic data as well as real financial data.
翻译:这项工作研究功能线性模型框架下的转让学习问题,其目的是通过利用相关来源模型的知识来改进目标模型的适切性。我们利用Recing Kernel Hilbert Spaces测量目标模型与源模型之间的关系,允许空间结构解释正在转让的知识类型。提出了两种算法:一种是在知道可转移来源指数时转让知识,而另一种是在没有事先获得来源信息的情况下利用汇总实现知识转让。此外,我们建立了超风险最佳趋同率,通过转让学习数学可实现统计收益。提议的算法的有效性在合成数据以及实际金融数据上得到证明。