Instruction-Oriented Preference Alignment for Enhancing Multi-Modal Comprehension Capability of MLLMs
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914023651409920 |
|---|---|
| author | Wang, Zitian Liao, Yue Rong, Kang Rao, Fengyun Yang, Yibo Liu, Si |
| author_facet | Wang, Zitian Liao, Yue Rong, Kang Rao, Fengyun Yang, Yibo Liu, Si |
| contents | Preference alignment has emerged as an effective strategy to enhance the performance of Multimodal Large Language Models (MLLMs) following supervised fine-tuning. While existing preference alignment methods predominantly target hallucination factors, they overlook the factors essential for multi-modal comprehension capabilities, often narrowing their improvements on hallucination mitigation. To bridge this gap, we propose Instruction-oriented Preference Alignment (IPA), a scalable framework designed to automatically construct alignment preferences grounded in instruction fulfillment efficacy. Our method involves an automated preference construction coupled with a dedicated verification process that identifies instruction-oriented factors, avoiding significant variability in response representations. Additionally, IPA incorporates a progressive preference collection pipeline, further recalling challenging samples through model self-evolution and reference-guided refinement. Experiments conducted on Qwen2VL-7B demonstrate IPA's effectiveness across multiple benchmarks, including hallucination evaluation, visual question answering, and text understanding tasks, highlighting its capability to enhance general comprehension. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_20309 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Instruction-Oriented Preference Alignment for Enhancing Multi-Modal Comprehension Capability of MLLMs Wang, Zitian Liao, Yue Rong, Kang Rao, Fengyun Yang, Yibo Liu, Si Computer Vision and Pattern Recognition Preference alignment has emerged as an effective strategy to enhance the performance of Multimodal Large Language Models (MLLMs) following supervised fine-tuning. While existing preference alignment methods predominantly target hallucination factors, they overlook the factors essential for multi-modal comprehension capabilities, often narrowing their improvements on hallucination mitigation. To bridge this gap, we propose Instruction-oriented Preference Alignment (IPA), a scalable framework designed to automatically construct alignment preferences grounded in instruction fulfillment efficacy. Our method involves an automated preference construction coupled with a dedicated verification process that identifies instruction-oriented factors, avoiding significant variability in response representations. Additionally, IPA incorporates a progressive preference collection pipeline, further recalling challenging samples through model self-evolution and reference-guided refinement. Experiments conducted on Qwen2VL-7B demonstrate IPA's effectiveness across multiple benchmarks, including hallucination evaluation, visual question answering, and text understanding tasks, highlighting its capability to enhance general comprehension. |
| title | Instruction-Oriented Preference Alignment for Enhancing Multi-Modal Comprehension Capability of MLLMs |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2503.20309 |