Year: 2024
Type: article
Abstract: This manuscript delves into the origin, progression, and future of multimodal fusion by conducting a comprehensive review of seminal research at various phases of its ontogeny. It encompasses three foundational methodologies of multimodal integration: Early Fusion, late Fusion, and Hybrid Fusion. Moreover, the article presents three novel methodologies for integration. It culminates in a discourse on potential subsequent applications, obstacles, and progress within the domain. solve the fusion problem of heterogeneous neural networks, and improve the correlation and consistency between modes in the aspects of cross-modal representation learning, multi-modal sentiment analysis, multi-modal intelligent interaction, so as to achieve better human-computer interaction experience.
DOI: 10.54097/sx342m55
Authors: Haowen Xue, Zhitao Zhu
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4405394465