Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Shan, Xiaojun, Cao, Qi, Han, Xing, Yu, Haofei, Liang, Paul Pu
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2506.02308
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909641417424896
author Shan, Xiaojun
Cao, Qi
Han, Xing
Yu, Haofei
Liang, Paul Pu
author_facet Shan, Xiaojun
Cao, Qi
Han, Xing
Yu, Haofei
Liang, Paul Pu
contents Recent advances in multimodal foundation models have achieved state-of-the-art performance across a range of tasks. These breakthroughs are largely driven by new pre-training paradigms that leverage large-scale, unlabeled multimodal data, followed by instruction fine-tuning on curated labeled datasets and high-quality prompts. While there is growing interest in scaling instruction fine-tuning to ever-larger datasets in both quantity and scale, our findings reveal that simply increasing the number of instruction-tuning tasks does not consistently yield better performance. Instead, we observe that grouping tasks by the common interactions across modalities, such as discovering redundant shared information, prioritizing modality selection with unique information, or requiring synergistic fusion to discover new information from both modalities, encourages the models to learn transferrable skills within a group while suppressing interference from mismatched tasks. To this end, we introduce MINT, a simple yet surprisingly effective task-grouping strategy based on the type of multimodal interaction. We demonstrate that the proposed method greatly outperforms existing task grouping baselines for multimodal instruction tuning, striking an effective balance between generalization and specialization.
format Preprint
id arxiv_https___arxiv_org_abs_2506_02308
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MINT: Multimodal Instruction Tuning with Multimodal Interaction Grouping
Shan, Xiaojun
Cao, Qi
Han, Xing
Yu, Haofei
Liang, Paul Pu
Machine Learning
Artificial Intelligence
Recent advances in multimodal foundation models have achieved state-of-the-art performance across a range of tasks. These breakthroughs are largely driven by new pre-training paradigms that leverage large-scale, unlabeled multimodal data, followed by instruction fine-tuning on curated labeled datasets and high-quality prompts. While there is growing interest in scaling instruction fine-tuning to ever-larger datasets in both quantity and scale, our findings reveal that simply increasing the number of instruction-tuning tasks does not consistently yield better performance. Instead, we observe that grouping tasks by the common interactions across modalities, such as discovering redundant shared information, prioritizing modality selection with unique information, or requiring synergistic fusion to discover new information from both modalities, encourages the models to learn transferrable skills within a group while suppressing interference from mismatched tasks. To this end, we introduce MINT, a simple yet surprisingly effective task-grouping strategy based on the type of multimodal interaction. We demonstrate that the proposed method greatly outperforms existing task grouping baselines for multimodal instruction tuning, striking an effective balance between generalization and specialization.
title MINT: Multimodal Instruction Tuning with Multimodal Interaction Grouping
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2506.02308