인문학
사회과학
자연과학
공학
의약학
농수해양학
예술체육학
복합학
DBpia AI
AI 에이전트
AI 아이디어
AI 리더
AI 뷰어
Citeasy
크롬 서지관리 다운로드
지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
피인용 0
초록·키워드
In Knowledge-Based Visual Question Answering (KB-VQA), which requires external knowledge for accurate question answering, significant advancements have been made through the use of Large Language Models (LLMs). BLIP-2, a popular multimodal LLM, employs a single-layer Q-Former for visual feature extraction and cross-modal interactions but faces challenges with complex reasoning tasks.
To overcome these limitations, we propose integrating the Multimodal Co-Attention Network (MCAN), which uses a multi-layered approach to enhance the interaction between visual and textual inputs. Additionally, we introduce Question-Aware Prompts during fine-tuning, combining Answer Candidates with confidence scores and Answer-Aware Examples from past cases. This improves the model's ability to interpret questions accurately and generate more contextually appropriate answers.
Experimental results on KB-VQA datasets show a 6.9% improvement in accuracy compared to baseline models, demonstrating the effectiveness of our approach in handling complex multimodal reasoning tasks.
To overcome these limitations, we propose integrating the Multimodal Co-Attention Network (MCAN), which uses a multi-layered approach to enhance the interaction between visual and textual inputs. Additionally, we introduce Question-Aware Prompts during fine-tuning, combining Answer Candidates with confidence scores and Answer-Aware Examples from past cases. This improves the model's ability to interpret questions accurately and generate more contextually appropriate answers.
Experimental results on KB-VQA datasets show a 6.9% improvement in accuracy compared to baseline models, demonstrating the effectiveness of our approach in handling complex multimodal reasoning tasks.
본문·목차
인공지능 문자 인식 모델을 통해 추출된 텍스트로, 일부 오타나 오류가 포함될 수 있으나 지속적으로 개선 중입니다.
오류를 발견하셨다면 해당 부분을 드래그한 후 ' 를 통해 신고해주세요.
오류를 발견하셨다면 해당 부분을 드래그한 후 ' 를 통해 신고해주세요.
최근 본 자료 전체보기
