Datasets:
license: mit
task_categories:
- feature-extraction
✨ Overview
This repository contains data and code for CalMRL, a framework designed to calibrate incomplete alignments caused by missing modalities in multimodal representation learning.
Multimodal representation learning harmonizes distinct modalities by aligning them into a unified latent space. CalMRL leverages priors and inherent connections among modalities to model the imputation for missing ones at the representation level, addressing the "anchor shift" problem where observed modalities align with local anchors that deviate from the optimal ones.
For more details, please refer to the paper: Calibrated Multimodal Representation Learning with Missing Modalities.
🎯 Key Features
- Calibration of Incomplete Alignments: Models imputation for missing modalities at the representation level.
- Anchor Shift Mitigation: Provides theoretical insights into how missing modalities cause local anchors to deviate from optimal ones.
- Bi-step Learning: Employs a bi-step learning method with a closed-form solution for the posterior distribution of shared latents.
- Flexibility: Offers the ability to absorb data with missing modalities by equipping calibrated alignment with existing advanced methods.
🏗️ Related Work: PMRL
This work builds upon or relates to Principled Multimodal Representation Learning (PMRL). PMRL addresses fundamental challenges in multimodal representation learning by proposing a framework that achieves simultaneous alignment of multiple modalities without anchor dependency.
Citation
If you find this work useful, please consider citing:
@article{liu2025calibrated,
title={Calibrated Multimodal Representation Learning with Missing Modalities},
author={Liu, Xiaohao and Xia, Xiaobo and Wei, Jiaheng and Yang, Shuo and Su, Xiu and Ng, See-Kiong and Chua, Tat-Seng},
journal={arXiv preprint arXiv:2511.12034},
year={2025}
}
@article{liu2026principled,
title={Principled multimodal representation learning},
author={Liu, Xiaohao and Xia, Xiaobo and Ng, See-Kiong and Chua, Tat-Seng},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
year={2026},
publisher={IEEE}
}