摘要
This paper addresses the distributed online bandit linear regression problems with privacy protection, in which the training data are spread in a multi-agent network. Each node identifies a linear predictor to fit the training data and experiences a square loss on each round. The purpose is to minimize the regret that assesses the difference of the accumulated loss between the online linear predictor and the optimal offline linear predictor. Moreover, the differential privacy strategy is adopted to prevent the adversary from inferring the parameter vector of any node. Two efficient differentially private distributed online regression algorithms are developed in the cases of one-point and two-point bandit feedback. Our upper bounds in O(K3 / 4 ) and O(root K) for one-point and two-point bandit feedback, respectively, where analysis suggests that the developed algorithms achieve e-differential privacy and establish the regret K is the time horizon. We also show that there exists a tradeoff between our algorithms' privacy level and convergence. Finally, the performance of the proposed algorithms is validated by a numerical example.
学科与关键词
- Privacy-Preserving Communication-Efficient Federated Multi-Armed BanditsLi, Tan*; Song, Linqi*IEEE Journal on Selected Areas in Communications, 2022, 40(3): 773-787.被引 8
- Distributed online bandit optimization under random quantizationYuan, Deming*; Zhang, Baoyong; Ho, Daniel W. C.; Zheng, Wei Xing; Xu, ShengyuanAutomatica, 2022, 146: 110590.被引 6
- Killing two birds with one stone: Quantization achieves privacy in distributed learningYan, Guangfeng; Li, Tan; Wu, Kui; Song, Linqi*Digital Signal Processing: A Review Journal , 2024, 146: 104353.
