Rejection via Learning Density Ratios

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Soen, Alexander, Husain, Hisham, Schulz, Philip, Nguyen, Vu
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915276614795264
author Soen, Alexander
Husain, Hisham
Schulz, Philip
Nguyen, Vu
author_facet Soen, Alexander
Husain, Hisham
Schulz, Philip
Nguyen, Vu
contents Classification with rejection emerges as a learning paradigm which allows models to abstain from making predictions. The predominant approach is to alter the supervised learning pipeline by augmenting typical loss functions, letting model rejection incur a lower loss than an incorrect prediction. Instead, we propose a different distributional perspective, where we seek to find an idealized data distribution which maximizes a pretrained model's performance. This can be formalized via the optimization of a loss's risk with a $φ$-divergence regularization term. Through this idealized distribution, a rejection decision can be made by utilizing the density ratio between this distribution and the data distribution. We focus on the setting where our $φ$-divergences are specified by the family of $α$-divergence. Our framework is tested empirically over clean and noisy datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2405_18686
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Rejection via Learning Density Ratios
Soen, Alexander
Husain, Hisham
Schulz, Philip
Nguyen, Vu
Machine Learning
Classification with rejection emerges as a learning paradigm which allows models to abstain from making predictions. The predominant approach is to alter the supervised learning pipeline by augmenting typical loss functions, letting model rejection incur a lower loss than an incorrect prediction. Instead, we propose a different distributional perspective, where we seek to find an idealized data distribution which maximizes a pretrained model's performance. This can be formalized via the optimization of a loss's risk with a $φ$-divergence regularization term. Through this idealized distribution, a rejection decision can be made by utilizing the density ratio between this distribution and the data distribution. We focus on the setting where our $φ$-divergences are specified by the family of $α$-divergence. Our framework is tested empirically over clean and noisy datasets.
title Rejection via Learning Density Ratios
topic Machine Learning
url https://arxiv.org/abs/2405.18686