作者位次 #3/4ACL 2026Findings· Findings of ACL
B - APO : Bias-Targeted Adversarial Preference Optimization for Debiasing Multimodal Large Language Models
Pinlong Zhao · Zike Ding · Zengshu Ye · Zhou Zhaoting
摘要
Multimodal Large Language Models (MLLMs) often suffer from modality bias, where the model disproportionately relies on one modality while neglecting critical information from others. Existing debiasing methods via modality masking create biased responses by completely removing an entire modality,...