From Hindsight to Foresight: Self-Encouraged Hindsight Distillation for Knowledge-based Visual Question Answering
从回顾到前瞻:面向知识驱动视觉问答的自我鼓励回顾蒸馏
机构 * College of Computing and Data Science, Nanyang Technological University(南洋理工大学计算与数据科学学院)
专题命中 知识库问答 :retrieval augmented generation(abstract)
AI总结 本文提出HinD框架,通过知识鼓励偏好优化提升多模态大语言模型的知识推理能力,实验表明其在OK-VQA和A-OKVQA上表现优异。