Abstract
This paper focuses on online kernel learning over a decentralized network.Each agent in the network receives continuous streaming data locally and workscollaboratively to learn a nonlinear prediction function that is globallyoptimal in the reproducing kernel Hilbert space with respect to the totalinstantaneous costs of all agents. In order to circumvent the curse ofdimensionality issue in traditional online kernel learning, we utilize randomfeature (RF) mapping to convert the non-parametric kernel learning problem intoa fixed-length parametric one in the RF space. We then propose a novel learningframework named Online Decentralized Kernel learning via Linearized ADMM(ODKLA) to efficiently solve the online decentralized kernel learning problem.To further improve the communication efficiency, we add the quantization andcensoring strategies in the communication stage and develop the Quantized andCommunication-censored ODKLA (QC-ODKLA) algorithm. We theoretically prove thatboth ODKLA and QC-ODKLA can achieve the optimal sublinear regret$\mathcal{O}(\sqrt{T})$ over $T$ time slots. Through numerical experiments, weevaluate the learning effectiveness, communication, and computationefficiencies of the proposed methods.