The Neural Testbed: Evaluating Joint Predictions
Abstract
Predictive distributions quantify uncertainties ignored by point estimates. This paper introduces The Neural Testbed: an open-source benchmark for controlled and principled evaluation of agents that generate such predictions. Crucially, the testbed assesses agents not only on the quality of their marginal predictions per input, but also on their joint predictions across many inputs. We evaluate a range of agents using a simple neural network data generating process. Our results indicate that some popular Bayesian deep learning agents do not fare well with joint predictions, even when they can produce accurate marginal predictions. We also show that the quality of joint predictions drives performance in downstream decision tasks. We find these results are robust across choice a wide range of generative models, and highlight the practical importance of joint predictions to the community.
Cite
@article{arxiv.2110.04629,
title = {The Neural Testbed: Evaluating Joint Predictions},
author = {Ian Osband and Zheng Wen and Seyed Mohammad Asghari and Vikranth Dwaracherla and Botao Hao and Morteza Ibrahimi and Dieterich Lawson and Xiuyuan Lu and Brendan O'Donoghue and Benjamin Van Roy},
journal= {arXiv preprint arXiv:2110.04629},
year = {2022}
}