一作ACL 2026Findings· Findings of ACL
Sem- DPO : Mitigating Semantic Inconsistency in Preference Optimization for Prompt Engineering
Anas Mohamed · Azal Ahmad Khan · Xinran Wang · Ahmad Faraz Khan · Shuwen Ge · Saman Bahzad Khan · Ayaan Ahmad · Ali Anwar
摘要
Generative AI can now synthesize strikingly realistic images from text, yet output quality remains highly sensitive to how prompts are phrased. Direct Preference Optimization (DPO) offers a lightweight, off-policy alternative to RL for automatic prompt engineering, but its token-level regularizat...