Minimax Learning for Remote Prediction
Abstract
The classical problem of supervised learning is to infer an accurate predictor of a target variable from a measured variable by using a finite number of labeled training samples. Motivated by the increasingly distributed nature of data and decision making, in this paper we consider a variation of this classical problem in which the prediction is performed remotely based on a rate-constrained description of . Upon receiving , the remote node computes an estimate of . We follow the recent minimax approach to study this learning problem and show that it corresponds to a one-shot minimax noisy source coding problem. We then establish information theoretic bounds on the risk-rate Lagrangian cost and a general method to design a near-optimal descriptor-estimator pair, which can be viewed as a rate-constrained analog to the maximum conditional entropy principle used in the classical minimax learning problem. Our results show that a naive estimate-compress scheme for rate-constrained prediction is not in general optimal.
Cite
@article{arxiv.1806.00071,
title = {Minimax Learning for Remote Prediction},
author = {Cheuk Ting Li and Xiugang Wu and Ayfer Ozgur and Abbas El Gamal},
journal= {arXiv preprint arXiv:1806.00071},
year = {2021}
}
Comments
10 pages, 4 figures, presented in part at ISIT 2018