Mixability of integral losses: A key to efficient online aggregation of functional and probabilistic forecasts

作者：

Highlights：

• Integral extensions of mixable loss functions are also mixable.

• Most of losses on probability distributions are mixable integral losses.

• Constant-regret online learning is possible with probabilistic forecasts.

摘要

•Integral extensions of mixable loss functions are also mixable.•Most of losses on probability distributions are mixable integral losses.•Constant-regret online learning is possible with probabilistic forecasts.

论文关键词：Integral loss functions,Mixability,Exponential concavity,Prediction with expert advice,Functional forecasting,Probabilistic forecasting

论文评审过程：Received 18 September 2020, Revised 10 May 2021, Accepted 4 June 2021, Available online 15 July 2021, Version of Record 22 July 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108175