APEX: Learning Adaptive Priorities for Multi-Objective Alignment in Vision-Language Generation
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914245370707968 |
|---|---|
| author | Chen, Dongliang Zhuang, Xinlin Xu, Junjie Xie, Luojian Wang, Zehui Zhuang, Jiaxi Yang, Haolin Dou, Liang He, Xiao Wu, Xingjiao Qian, Ying |
| author_facet | Chen, Dongliang Zhuang, Xinlin Xu, Junjie Xie, Luojian Wang, Zehui Zhuang, Jiaxi Yang, Haolin Dou, Liang He, Xiao Wu, Xingjiao Qian, Ying |
| contents | Multi-objective alignment for text-to-image generation is commonly implemented via static linear scalarization, but fixed weights often fail under heterogeneous rewards, leading to optimization imbalance where models overfit high-variance, high-responsiveness objectives (e.g., OCR) while under-optimizing perceptual goals. We identify two mechanistic causes: variance hijacking, where reward dispersion induces implicit reweighting that dominates the normalized training signal, and gradient conflicts, where competing objectives produce opposing update directions and trigger seesaw-like oscillations. We propose APEX (Adaptive Priority-based Efficient X-objective Alignment), which stabilizes heterogeneous rewards with Dual-Stage Adaptive Normalization and dynamically schedules objectives via P^3 Adaptive Priorities that combine learning potential, conflict penalty, and progress need. On Stable Diffusion 3.5, APEX achieves improved Pareto trade-offs across four heterogeneous objectives, with balanced gains of +1.31 PickScore, +0.35 DeQA, and +0.53 Aesthetics while maintaining competitive OCR accuracy, mitigating the instability of multi-objective alignment. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_06574 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | APEX: Learning Adaptive Priorities for Multi-Objective Alignment in Vision-Language Generation Chen, Dongliang Zhuang, Xinlin Xu, Junjie Xie, Luojian Wang, Zehui Zhuang, Jiaxi Yang, Haolin Dou, Liang He, Xiao Wu, Xingjiao Qian, Ying Computer Vision and Pattern Recognition Multi-objective alignment for text-to-image generation is commonly implemented via static linear scalarization, but fixed weights often fail under heterogeneous rewards, leading to optimization imbalance where models overfit high-variance, high-responsiveness objectives (e.g., OCR) while under-optimizing perceptual goals. We identify two mechanistic causes: variance hijacking, where reward dispersion induces implicit reweighting that dominates the normalized training signal, and gradient conflicts, where competing objectives produce opposing update directions and trigger seesaw-like oscillations. We propose APEX (Adaptive Priority-based Efficient X-objective Alignment), which stabilizes heterogeneous rewards with Dual-Stage Adaptive Normalization and dynamically schedules objectives via P^3 Adaptive Priorities that combine learning potential, conflict penalty, and progress need. On Stable Diffusion 3.5, APEX achieves improved Pareto trade-offs across four heterogeneous objectives, with balanced gains of +1.31 PickScore, +0.35 DeQA, and +0.53 Aesthetics while maintaining competitive OCR accuracy, mitigating the instability of multi-objective alignment. |
| title | APEX: Learning Adaptive Priorities for Multi-Objective Alignment in Vision-Language Generation |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2601.06574 |