Research map / Vision-language and multimodal / Cross-modal learning
Calibrated Multimodal Representation Learning with Missing Modalities
Xiaohao Liu, Xiaobo Xia, Jiaheng Wei, Shuo Yang, Xiu Su, See-Kiong Ng, Tat-Seng Chua
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| impute missing modalities at representation level with generative model (CalMRL) on top of PMRL | PMRL with VAST backbone · 6 retrieval benchmarks, avg Recall@1 | avg Recall@1: 53.8 -> 54.2 | Improved |
| add CalMRL calibration for classification | PMRL with ImageBind backbone · VGGSound, UCF101, AudioSet, ESC50 avg accuracy | accuracy: 44.04 -> 45.19 | Improved |
| continual training on missing-modality datasets | VAST · 6 retrieval benchmarks, avg Recall@1 | avg Recall@1: 44.8 -> 53.3 | Improved |
| train on sole audio-text dataset without calibration (ablation) | VAST · DiDeMo T→V and ActivityNet T→V | Recall@1: -3.6 and -2.9 vs VAST | Hurt |
| add CalMRL calibration on sole audio-text training (ablation) | PMRL with sole audio-text training · 6 retrieval benchmarks, avg Recall@1 | avg Recall@1: 51.2 -> 51.7 | Improved |
| add CalMRL calibration on sole video-text training (ablation) | PMRL with sole video-text training · 6 retrieval benchmarks, avg Recall@1 | avg Recall@1: 50.0 -> 50.4 | Improved |
| replace linear-Gaussian generative imputation with MVAE (ablation) | CalMRL · 6 retrieval benchmarks, avg Recall@1 | avg Recall@1: 54.8 -> 54.0 | Hurt |
| replace generative imputation with MoPoE (ablation) | CalMRL · 6 retrieval benchmarks, avg Recall@1 | avg Recall@1: 54.8 -> 54.1 | Hurt |
| replace generative imputation with SMIL or Knowledge Bridger (ablation) | CalMRL · 6 retrieval benchmarks, avg Recall@1 | avg Recall@1: 54.8 -> 54.0 and 54.2 | Hurt |
| remove instance matching loss (alpha=0) (ablation) | CalMRL · 6 retrieval benchmarks, avg Recall@1 | avg Recall@1: 54.80 -> 53.30 | Hurt |
| over-weight instance matching loss (alpha=1) (ablation) | CalMRL · 6 retrieval benchmarks, avg Recall@1 | avg Recall@1: 54.80 -> 53.21 | Hurt |
| vary temperature tau from 0.05 to 0.1-1 (ablation) | CalMRL · 6 retrieval benchmarks, avg Recall@1 | avg Recall@1: 54.80 -> 54.37 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VAST↑ — VAST: A Vision-Audio-Subtitle-Text Omni-Modality Foundation Model and Dataset (NeurIPS 2023) · better on 2 of 2 shared results
- ImageBind — ImageBind: One Embedding Space To Bind Them All (CVPR 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- On the Value of Cross-Modal Misalignment in Multimodal Representation Learning (NeurIPS 2025 · similarity 0.90)
- With Limited Data for Multimodal Alignment, Let the STRUCTURE Guide You (NeurIPS 2025 · similarity 0.89)
- Aligning Multimodal Representations through an Information Bottleneck (ICML 2025 · similarity 0.89)
- Unity by Diversity: Improved Representation Learning for Multimodal VAEs (NeurIPS 2024 · similarity 0.89)
- Do More Modalities Always Help? A Geometric Perspective on Missing-Modality Robustness (NeurIPS 2026 · similarity 0.89)
- UNIALIGN: Scaling Multimodal Alignment within One Unified Model (CVPR 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
