Guardado en:
Detalles Bibliográficos
Autores principales: Vassef, Shayan, Dabiriaghdam, Amirhossein, Bakhtiari, Mohammadreza, Yaghoobzadeh, Yadollah
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2507.15236
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911066860027904
author Vassef, Shayan
Dabiriaghdam, Amirhossein
Bakhtiari, Mohammadreza
Yaghoobzadeh, Yadollah
author_facet Vassef, Shayan
Dabiriaghdam, Amirhossein
Bakhtiari, Mohammadreza
Yaghoobzadeh, Yadollah
contents This work investigates the impact of multi-task, multi-lingual, and multi-source learning approaches on the robustness and performance of pretrained language models. To enhance this analysis, we introduce Subsets of Interest (SOI), a novel categorization framework that identifies six distinct learning behavior patterns during training, including forgettable examples, unlearned examples, and always correct examples. Through SOI transition heatmaps and dataset cartography visualization, we analyze how examples shift between these categories when transitioning from single-setting to multi-setting configurations. We perform comprehensive experiments across three parallel comparisons: multi-task vs. single-task learning using English tasks (entailment, paraphrase, sentiment), multi-source vs. single-source learning using sentiment analysis datasets, and multi-lingual vs. single-lingual learning using intent classification in French, English, and Persian. Our results demonstrate that multi-source learning consistently improves out-of-distribution performance by up to 7%, while multi-task learning shows mixed results with notable gains in similar task combinations. We further introduce a two-stage fine-tuning approach where the second stage leverages SOI-based subset selection to achieve additional performance improvements. These findings provide new insights into training dynamics and offer practical approaches for optimizing multi-setting language model performance.
format Preprint
id arxiv_https___arxiv_org_abs_2507_15236
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SOI Matters: Analyzing Multi-Setting Training Dynamics in Pretrained Language Models via Subsets of Interest
Vassef, Shayan
Dabiriaghdam, Amirhossein
Bakhtiari, Mohammadreza
Yaghoobzadeh, Yadollah
Computation and Language
Machine Learning
This work investigates the impact of multi-task, multi-lingual, and multi-source learning approaches on the robustness and performance of pretrained language models. To enhance this analysis, we introduce Subsets of Interest (SOI), a novel categorization framework that identifies six distinct learning behavior patterns during training, including forgettable examples, unlearned examples, and always correct examples. Through SOI transition heatmaps and dataset cartography visualization, we analyze how examples shift between these categories when transitioning from single-setting to multi-setting configurations. We perform comprehensive experiments across three parallel comparisons: multi-task vs. single-task learning using English tasks (entailment, paraphrase, sentiment), multi-source vs. single-source learning using sentiment analysis datasets, and multi-lingual vs. single-lingual learning using intent classification in French, English, and Persian. Our results demonstrate that multi-source learning consistently improves out-of-distribution performance by up to 7%, while multi-task learning shows mixed results with notable gains in similar task combinations. We further introduce a two-stage fine-tuning approach where the second stage leverages SOI-based subset selection to achieve additional performance improvements. These findings provide new insights into training dynamics and offer practical approaches for optimizing multi-setting language model performance.
title SOI Matters: Analyzing Multi-Setting Training Dynamics in Pretrained Language Models via Subsets of Interest
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2507.15236