매일 아침, 어제의 AI를 한 통으로 정리해 보내드립니다메일로 받아보기

METAL LAB

MIDAS: Mutual Information Disentanglement with Uncertainty-Aware Fusion for Incomplete Multimodal Sentiment Analysis

arXiv:2608.099862026-08-12

arXiv:2608.09986v1 Announce Type: new Abstract: Most existing multimodal sentiment analysis approaches assume access to complete multimodal inputs. However, real-world applications frequently encounter incomplete or corrupted modalities, posing a critical challenge. Although several methods have been proposed to tackle this issue, they mainly rely on data imputation and heuristic coordination constraints, which fail to effectively extract and leverage task-relevant information from the incomplet

저자 · Yuhua Wen, Yingying Zhou, Qifei Li, Yingming Gao, Zhengqi Wen, Jianhua Tao, Ya Li

arXiv에서 원문 보기