Early-Exit Neural Networks with Nested Prediction Sets

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jazbec, Metod, Forré, Patrick, Mandt, Stephan, Zhang, Dan, Nalisnick, Eric
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929370685243392
author Jazbec, Metod
Forré, Patrick
Mandt, Stephan
Zhang, Dan
Nalisnick, Eric
author_facet Jazbec, Metod
Forré, Patrick
Mandt, Stephan
Zhang, Dan
Nalisnick, Eric
contents Early-exit neural networks (EENNs) enable adaptive and efficient inference by providing predictions at multiple stages during the forward pass. In safety-critical applications, these predictions are meaningful only when accompanied by reliable uncertainty estimates. A popular method for quantifying the uncertainty of predictive models is the use of prediction sets. However, we demonstrate that standard techniques such as conformal prediction and Bayesian credible sets are not suitable for EENNs. They tend to generate non-nested sets across exits, meaning that labels deemed improbable at one exit may reappear in the prediction set of a subsequent exit. To address this issue, we investigate anytime-valid confidence sequences (AVCSs), an extension of traditional confidence intervals tailored for data-streaming scenarios. These sequences are inherently nested and thus well-suited for an EENN's sequential predictions. We explore the theoretical and practical challenges of using AVCSs in EENNs and show that they indeed yield nested sets across exits. Thus our work presents a promising approach towards fast, yet still safe, predictive modeling
format Preprint
id arxiv_https___arxiv_org_abs_2311_05931
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Early-Exit Neural Networks with Nested Prediction Sets
Jazbec, Metod
Forré, Patrick
Mandt, Stephan
Zhang, Dan
Nalisnick, Eric
Machine Learning
Artificial Intelligence
Early-exit neural networks (EENNs) enable adaptive and efficient inference by providing predictions at multiple stages during the forward pass. In safety-critical applications, these predictions are meaningful only when accompanied by reliable uncertainty estimates. A popular method for quantifying the uncertainty of predictive models is the use of prediction sets. However, we demonstrate that standard techniques such as conformal prediction and Bayesian credible sets are not suitable for EENNs. They tend to generate non-nested sets across exits, meaning that labels deemed improbable at one exit may reappear in the prediction set of a subsequent exit. To address this issue, we investigate anytime-valid confidence sequences (AVCSs), an extension of traditional confidence intervals tailored for data-streaming scenarios. These sequences are inherently nested and thus well-suited for an EENN's sequential predictions. We explore the theoretical and practical challenges of using AVCSs in EENNs and show that they indeed yield nested sets across exits. Thus our work presents a promising approach towards fast, yet still safe, predictive modeling
title Early-Exit Neural Networks with Nested Prediction Sets
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2311.05931