Fast yet Safe: Early-Exiting with Risk Control

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jazbec, Metod, Timans, Alexander, Veljković, Tin Hadži, Sakmann, Kaspar, Zhang, Dan, Naesseth, Christian A., Nalisnick, Eric
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915004330016768
author Jazbec, Metod
Timans, Alexander
Veljković, Tin Hadži
Sakmann, Kaspar
Zhang, Dan
Naesseth, Christian A.
Nalisnick, Eric
author_facet Jazbec, Metod
Timans, Alexander
Veljković, Tin Hadži
Sakmann, Kaspar
Zhang, Dan
Naesseth, Christian A.
Nalisnick, Eric
contents Scaling machine learning models significantly improves their performance. However, such gains come at the cost of inference being slow and resource-intensive. Early-exit neural networks (EENNs) offer a promising solution: they accelerate inference by allowing intermediate layers to exit and produce a prediction early. Yet a fundamental issue with EENNs is how to determine when to exit without severely degrading performance. In other words, when is it 'safe' for an EENN to go 'fast'? To address this issue, we investigate how to adapt frameworks of risk control to EENNs. Risk control offers a distribution-free, post-hoc solution that tunes the EENN's exiting mechanism so that exits only occur when the output is of sufficient quality. We empirically validate our insights on a range of vision and language tasks, demonstrating that risk control can produce substantial computational savings, all the while preserving user-specified performance goals.
format Preprint
id arxiv_https___arxiv_org_abs_2405_20915
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Fast yet Safe: Early-Exiting with Risk Control
Jazbec, Metod
Timans, Alexander
Veljković, Tin Hadži
Sakmann, Kaspar
Zhang, Dan
Naesseth, Christian A.
Nalisnick, Eric
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Scaling machine learning models significantly improves their performance. However, such gains come at the cost of inference being slow and resource-intensive. Early-exit neural networks (EENNs) offer a promising solution: they accelerate inference by allowing intermediate layers to exit and produce a prediction early. Yet a fundamental issue with EENNs is how to determine when to exit without severely degrading performance. In other words, when is it 'safe' for an EENN to go 'fast'? To address this issue, we investigate how to adapt frameworks of risk control to EENNs. Risk control offers a distribution-free, post-hoc solution that tunes the EENN's exiting mechanism so that exits only occur when the output is of sufficient quality. We empirically validate our insights on a range of vision and language tasks, demonstrating that risk control can produce substantial computational savings, all the while preserving user-specified performance goals.
title Fast yet Safe: Early-Exiting with Risk Control
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2405.20915