Learning Non-Parametric Surrogate Losses With Correlated Gradients

Training models by minimizing surrogate loss functions with gradient-based algorithms is a standard approach in various vision tasks. This strategy often leads to suboptimal solutions due to the gap between the target evaluation metrics and surrogate loss functions. In this paper, we propose a frame...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Seungdong Yoa, Jinyoung Park, Hyunwoo J. Kim
Formato:	article
Lenguaje:	EN
Publicado:	IEEE 2021
Materias:	Learning loss deep learning machine learning computer vision Electrical engineering. Electronics. Nuclear engineering TK1-9971
Acceso en línea:	https://doaj.org/article/ca7f6616a3d54055ac07358cd8af428b
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:ca7f6616a3d54055ac07358cd8af428b
record_format	dspace
spelling	oai:doaj.org-article:ca7f6616a3d54055ac07358cd8af428b2021-11-05T23:00:27ZLearning Non-Parametric Surrogate Losses With Correlated Gradients2169-353610.1109/ACCESS.2021.3120092https://doaj.org/article/ca7f6616a3d54055ac07358cd8af428b2021-01-01T00:00:00Zhttps://ieeexplore.ieee.org/document/9570322/https://doaj.org/toc/2169-3536Training models by minimizing surrogate loss functions with gradient-based algorithms is a standard approach in various vision tasks. This strategy often leads to suboptimal solutions due to the gap between the target evaluation metrics and surrogate loss functions. In this paper, we propose a framework to learn a surrogate loss function that approximates the evaluation metric with correlated gradients. We observe that the correlated gradients significantly benefit the gradient-based algorithms to improve the quality of solutions. We verify the effectiveness of our method in various tasks such as multi-class classification, ordinal regression, and pose estimation with three evaluation metrics and five datasets. Our extensive experiments showed that our method outperforms conventional loss functions and surrogate loss learning methods.Seungdong YoaJinyoung ParkHyunwoo J. KimIEEEarticleLearning lossdeep learningmachine learningcomputer visionElectrical engineering. Electronics. Nuclear engineeringTK1-9971ENIEEE Access, Vol 9, Pp 141199-141209 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	Learning loss deep learning machine learning computer vision Electrical engineering. Electronics. Nuclear engineering TK1-9971
spellingShingle	Learning loss deep learning machine learning computer vision Electrical engineering. Electronics. Nuclear engineering TK1-9971 Seungdong Yoa Jinyoung Park Hyunwoo J. Kim Learning Non-Parametric Surrogate Losses With Correlated Gradients
description	Training models by minimizing surrogate loss functions with gradient-based algorithms is a standard approach in various vision tasks. This strategy often leads to suboptimal solutions due to the gap between the target evaluation metrics and surrogate loss functions. In this paper, we propose a framework to learn a surrogate loss function that approximates the evaluation metric with correlated gradients. We observe that the correlated gradients significantly benefit the gradient-based algorithms to improve the quality of solutions. We verify the effectiveness of our method in various tasks such as multi-class classification, ordinal regression, and pose estimation with three evaluation metrics and five datasets. Our extensive experiments showed that our method outperforms conventional loss functions and surrogate loss learning methods.
format	article
author	Seungdong Yoa Jinyoung Park Hyunwoo J. Kim
author_facet	Seungdong Yoa Jinyoung Park Hyunwoo J. Kim
author_sort	Seungdong Yoa
title	Learning Non-Parametric Surrogate Losses With Correlated Gradients
title_short	Learning Non-Parametric Surrogate Losses With Correlated Gradients
title_full	Learning Non-Parametric Surrogate Losses With Correlated Gradients
title_fullStr	Learning Non-Parametric Surrogate Losses With Correlated Gradients
title_full_unstemmed	Learning Non-Parametric Surrogate Losses With Correlated Gradients
title_sort	learning non-parametric surrogate losses with correlated gradients
publisher	IEEE
publishDate	2021
url	https://doaj.org/article/ca7f6616a3d54055ac07358cd8af428b
work_keys_str_mv	AT seungdongyoa learningnonparametricsurrogatelosseswithcorrelatedgradients AT jinyoungpark learningnonparametricsurrogatelosseswithcorrelatedgradients AT hyunwoojkim learningnonparametricsurrogatelosseswithcorrelatedgradients
_version_	1718444008170061824

Learning Non-Parametric Surrogate Losses With Correlated Gradients

Ejemplares similares