The authors consider the online prediction from experts problem under differential privacy.
Differentially private online prediction achieves regret proportional to square root of best expert's loss
Authors propose the first private algorithm that adapts to low-loss regimes, nearly matching non-private optimal bounds.
Top University
Bo Li · Peng Ye
Guangzhou HKUST Fok Ying Tung Research Institute · HKUST
Research Digest··3 min read
Bo Li and Peng Ye present the first differentially private algorithm for online prediction from experts that attains regret Õ(√L*), where L* is the cumulative loss of the best expert.
§