Abstract
We obtain a lower bound for an algorithm predicting finite-dimensional distributions (i.e., points from a simplex) under Kullback-Leibler loss. The bound holds w.r.t.~the class of softmax linear predictors. We then show that the bound is asymptotically matched by the Bayesian universal algorithm.
Original language | English |
---|---|
Title of host publication | Conformal and Probabilistic Prediction and Applications 2021 |
Editors | Lars Carlsson, Zhiyuan Luo, Giovanni Cherubin, Khuong An Nguyen |
Publisher | Proceedings of Machine Learning Research |
Pages | 39-51 |
Number of pages | 13 |
Volume | 152 |
Publication status | Published - Sept 2021 |
Event | 10th Symposium on Conformal and Probabilistic Prediction with Applications, COPA 2021 - Online Duration: 8 Sept 2021 → 10 Sept 2021 Conference number: 10th |
Conference
Conference | 10th Symposium on Conformal and Probabilistic Prediction with Applications, COPA 2021 |
---|---|
Abbreviated title | COPA 2021 |
Period | 8/09/21 → 10/09/21 |
Keywords
- online learning
- competitive prediction
- loss bounds