Quantification of Predictive Uncertainty via Inference-Time Sampling

TL;DR

This paper introduces a post-hoc sampling method to estimate predictive uncertainty in deterministic models, enabling diverse, multi-modal predictions without altering the model architecture or training process.

Contribution

It presents a novel, architecture-agnostic sampling strategy for uncertainty estimation that works with any trained deterministic network, addressing overconfidence issues.

Findings

Generates diverse, multi-modal predictive distributions.

Uncertainty estimates correlate well with prediction errors.

Applicable to both imaging and non-imaging regression tasks.

Abstract

Predictive variability due to data ambiguities has typically been addressed via construction of dedicated models with built-in probabilistic capabilities that are trained to predict uncertainty estimates as variables of interest. These approaches require distinct architectural components and training mechanisms, may include restrictive assumptions and exhibit overconfidence, i.e., high confidence in imprecise predictions. In this work, we propose a post-hoc sampling strategy for estimating predictive uncertainty accounting for data ambiguity. The method can generate different plausible outputs for a given input and does not assume parametric forms of predictive distributions. It is architecture agnostic and can be applied to any feed-forward deterministic network without changes to the architecture or training procedure. Experiments on regression tasks on imaging and non-imaging input data show the method's ability to generate diverse and multi-modal predictive distributions, and a desirable correlation of the estimated uncertainty with the prediction error.