Robult: A Scalable Framework for Semi-Supervised Multimodal Learning with Missing Modalities

Duy A. Nguyen, Abhi Kamboj, Minh N. Do

Keywords: Multimodal learning, Semi-supervised learning, Missing modalities

TL;DR: This paper introduces Robult, a framework for robust multimodal learning that tackles the challenges of missing modalities and limited labeled data in semi-supervised settings.

Abstract: In multimodal learning, the presence of missing modalities and limited labeled data presents significant challenges for building robust models. We propose **Robult**, a novel framework designed to address these challenges by leveraging an information-theoretic approach to preserve modality-specific features and synergistic information across modalities. Our model introduces two key objectives: (1) a latent reconstruction loss to retain unique modality-specific information, and (2) a novel soft Positive-Unlabeled (PU) contrastive loss to efficiently utilize sparse labeled data in semi-supervised settings. Robult seamlessly integrates into deep learning architectures, enhancing performance across multiple downstream tasks and ensuring robustness even when modalities are missing at inference time. Empirical results across diverse datasets demonstrate that Robult surpasses existing methods in handling both semi-supervised learning and missing modalities, while its lightweight design enables scalability and easy integration with existing frameworks.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 8754