Unbiased Missing-Modality Multimodal Learning
Authors:
- 19 October 2025
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 24507-24517
Abstract
Recovering missing modalities in multimodal learning has recently been approached using diffusion models to synthesize absent data conditioned on available modalities. However, existing methods often suffer from modality generation bias: while certain modalities are generated with high fidelity, others—such as video—remain challenging due to intrinsic modality gaps, leading to imbalanced training. To address this issue, we propose $\boldsymbol{M} \boldsymbol{D}^{2} \boldsymbol{N}$ (Multi-stage Duplex Diffusion Network), a novel framework for unbiased missing-modality recovery. $M D^{2} N$ introduces a modality transfer module within a duplex diffusion architecture, enabling bidirectional generation between available and missing modalities through three stages: (1) global structure generation, (2) modality transfer, and (3) local crossmodal refinement. By training with duplex diffusion, both available and missing modalities generate each other in an intersecting manner, effectively achieving a balanced generation state. Extensive experiments demonstrate that $M D^{2} N$ significantly outperforms existing state-of-the-art methods, achieving up to 4% improvement over IMDer on the CMU-MOSEI dataset. Project page: here.
Keywords
Funding Information
- National Natural Science Foundation of China (62306064)
This publication has 47 references indexed in Scilit:
- Diffusion Models for Image Restoration and Enhancement: A Comprehensive SurveyInternational Journal of Computer Vision, 2025
- Towards Multimodal Sentiment Analysis Debiasing via Bias Purification2024
- Diffusion Probabilistic Model Assisted 3D Form Finding and Design Latent Space Exploration: A Case Study for Taihu Stone Spacial Transformation2024
- Transferable and differentiable discrete network embedding for multi-domains with hierarchical knowledge distillationInformation Sciences, 2023
- RETRACTED ARTICLE: ICDN: integrating consistency and difference networks by transformer for multimodal sentiment analysisApplied Intelligence, 2022
- Estimation of missing values in heterogeneous traffic data: Application of multimodal deep learning modelKnowledge-Based Systems, 2020
- A review: Deep learning for medical image segmentation using multi-modality fusionArray, 2019
- Degradation data analysis and remaining useful life estimation: A review on Wiener-process-based methodsEuropean Journal of Operational Research, 2018
- The truncated Euler–Maruyama method for stochastic differential equationsJournal of Computational and Applied Mathematics, 2015
- Reverse-time diffusion equation modelsStochastic Processes and their Applications, 1982