Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Neo, Dexter, Winkler, Stefan, Chen, Tsuhan
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2310.17159
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866929269349810176
author Neo, Dexter
Winkler, Stefan
Chen, Tsuhan
author_facet Neo, Dexter
Winkler, Stefan
Chen, Tsuhan
contents We present a new loss function that addresses the out-of-distribution (OOD) calibration problem. While many objective functions have been proposed to effectively calibrate models in-distribution, our findings show that they do not always fare well OOD. Based on the Principle of Maximum Entropy, we incorporate helpful statistical constraints observed during training, delivering better model calibration without sacrificing accuracy. We provide theoretical analysis and show empirically that our method works well in practice, achieving state-of-the-art calibration on both synthetic and real-world benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2310_17159
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle MaxEnt Loss: Constrained Maximum Entropy for Calibration under Out-of-Distribution Shift
Neo, Dexter
Winkler, Stefan
Chen, Tsuhan
Machine Learning
We present a new loss function that addresses the out-of-distribution (OOD) calibration problem. While many objective functions have been proposed to effectively calibrate models in-distribution, our findings show that they do not always fare well OOD. Based on the Principle of Maximum Entropy, we incorporate helpful statistical constraints observed during training, delivering better model calibration without sacrificing accuracy. We provide theoretical analysis and show empirically that our method works well in practice, achieving state-of-the-art calibration on both synthetic and real-world benchmarks.
title MaxEnt Loss: Constrained Maximum Entropy for Calibration under Out-of-Distribution Shift
topic Machine Learning
url https://arxiv.org/abs/2310.17159