Guardado en:
Detalles Bibliográficos
Autores principales: Kang, Junyong, Son, Donghyun, Song, Hwanjun, Chang, Buru
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:https://arxiv.org/abs/2411.19581
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909408554909696
author Kang, Junyong
Son, Donghyun
Song, Hwanjun
Chang, Buru
author_facet Kang, Junyong
Son, Donghyun
Song, Hwanjun
Chang, Buru
contents In-context learning refers to the emerging ability of large language models (LLMs) to perform a target task without additional training, utilizing demonstrations of the task. Recent studies aim to enhance in-context learning performance by selecting more useful demonstrations. However, they overlook the presence of inevitable noisy labels in task demonstrations that arise during the labeling process in the real-world. In this paper, we propose a new task, in-context learning with noisy labels, which aims to solve real-world problems for in-context learning where labels in task demonstrations would be corrupted. Moreover, we propose a new method and baseline methods for the new task, inspired by studies in learning with noisy labels. Through experiments, we demonstrate that our proposed method can serve as a safeguard against performance degradation in in-context learning caused by noisy labels.
format Preprint
id arxiv_https___arxiv_org_abs_2411_19581
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle In-Context Learning with Noisy Labels
Kang, Junyong
Son, Donghyun
Song, Hwanjun
Chang, Buru
Computation and Language
In-context learning refers to the emerging ability of large language models (LLMs) to perform a target task without additional training, utilizing demonstrations of the task. Recent studies aim to enhance in-context learning performance by selecting more useful demonstrations. However, they overlook the presence of inevitable noisy labels in task demonstrations that arise during the labeling process in the real-world. In this paper, we propose a new task, in-context learning with noisy labels, which aims to solve real-world problems for in-context learning where labels in task demonstrations would be corrupted. Moreover, we propose a new method and baseline methods for the new task, inspired by studies in learning with noisy labels. Through experiments, we demonstrate that our proposed method can serve as a safeguard against performance degradation in in-context learning caused by noisy labels.
title In-Context Learning with Noisy Labels
topic Computation and Language
url https://arxiv.org/abs/2411.19581