Guardado en:
Detalles Bibliográficos
Autores principales: Wang, Hongyi, Polo, Felipe Maia, Sun, Yuekai, Kundu, Souvik, Xing, Eric, Yurochkin, Mikhail
Formato: Preprint
Publicado: 2023
Materias:
Acceso en línea:https://arxiv.org/abs/2310.01542
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866916239650062336
author Wang, Hongyi
Polo, Felipe Maia
Sun, Yuekai
Kundu, Souvik
Xing, Eric
Yurochkin, Mikhail
author_facet Wang, Hongyi
Polo, Felipe Maia
Sun, Yuekai
Kundu, Souvik
Xing, Eric
Yurochkin, Mikhail
contents Training AI models that generalize across tasks and domains has long been among the open problems driving AI research. The emergence of Foundation Models made it easier to obtain expert models for a given task, but the heterogeneity of data that may be encountered at test time often means that any single expert is insufficient. We consider the Fusion of Experts (FoE) problem of fusing outputs of expert models with complementary knowledge of the data distribution and formulate it as an instance of supervised learning. Our method is applicable to both discriminative and generative tasks and leads to significant performance improvements in image and text classification, text summarization, multiple-choice QA, and automatic evaluation of generated text. We also extend our method to the "frugal" setting where it is desired to reduce the number of expert model evaluations at test time. Our implementation is publicly available at https://github.com/hwang595/FoE-ICLR2024.
format Preprint
id arxiv_https___arxiv_org_abs_2310_01542
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Fusing Models with Complementary Expertise
Wang, Hongyi
Polo, Felipe Maia
Sun, Yuekai
Kundu, Souvik
Xing, Eric
Yurochkin, Mikhail
Machine Learning
Training AI models that generalize across tasks and domains has long been among the open problems driving AI research. The emergence of Foundation Models made it easier to obtain expert models for a given task, but the heterogeneity of data that may be encountered at test time often means that any single expert is insufficient. We consider the Fusion of Experts (FoE) problem of fusing outputs of expert models with complementary knowledge of the data distribution and formulate it as an instance of supervised learning. Our method is applicable to both discriminative and generative tasks and leads to significant performance improvements in image and text classification, text summarization, multiple-choice QA, and automatic evaluation of generated text. We also extend our method to the "frugal" setting where it is desired to reduce the number of expert model evaluations at test time. Our implementation is publicly available at https://github.com/hwang595/FoE-ICLR2024.
title Fusing Models with Complementary Expertise
topic Machine Learning
url https://arxiv.org/abs/2310.01542