MMSP-L4.6

ATTENTION TO DETAILS, LOGITS TO TRUTH: VISUAL-AWARE ATTENTION AND LOGITS ENHANCEMENT TO MITIGATE HALLUCINATIONS IN LVLMS

Jingyi Wang, Fei Li, Rujie Liu, Fujitsu Research and Development Center, China
Session:
MMSP-L4: Factuality and Hallucination Mitigation in LMMs Oral
Track:
Multimedia Signal Processing [MM]
Location:
Room 131+132
Presentation Time:
Fri, 8 May, 10:40 - 11:00
Presentation
Discussion
Resources
No resources available.
Session MMSP-L4
MMSP-L4.1: Bifrost: An Adaptive Decision Framework for Regulating Depth of Thought in LLM Agents
Yanfang Zhou, Yuntao Liu, Academy of Military Sciences, China; Xiaodong Li, Jinlong Tian, National University of Defense Technology, China; Xinhai Xu, Academy of Military Sciences, China
MMSP-L4.2: CVSTIM: MITIGATING OBJECT HALLUCINATION IN MLLMS VIA CO-OCCURRENCE GUIDED VISUAL STIMULATION
Lan Wang, Ping Kuang, University of Electronic Science and Technology of China, China
MMSP-L4.3: Entropy-Aware Multimodal Preference Optimization for Factuality Alignment in Medical Visual Question Answering
Zhi Chen, Beiji Zou, Xiaoyan Kui, Central South University, China; Wenqi Lu, Manchester Metropolitan University, United Kingdom of Great Britain and Northern Ireland; Jinming Duan, University of Manchester, United Kingdom of Great Britain and Northern Ireland
MMSP-L4.4: SCHROMIND: MITIGATING HALLUCINATIONS IN MULTIMODAL LARGE LANGUAGE ¨ MODELS VIA SOLVING THE SCHRODINGER BRIDGE PROBLEM
Ziqiang Shi, Rujie Liu, Fujitsu Research & Development Center Co.,LTD., China; Shanshan Yu, Satoshi Munakata, Koichi Shirahata, Fujitsu Limited, Japan
MMSP-L4.5: MMFAST: RETHINKING VISION-LANGUAGE INTERACTION IN EFFICIENT MLLMS
shengyi xiong, independent researcher, Australia
MMSP-L4.6: ATTENTION TO DETAILS, LOGITS TO TRUTH: VISUAL-AWARE ATTENTION AND LOGITS ENHANCEMENT TO MITIGATE HALLUCINATIONS IN LVLMS
Jingyi Wang, Fei Li, Rujie Liu, Fujitsu Research and Development Center, China
Contacts