Research map / Label-efficient and robust learning / Self-supervised and contrastive
Generalized Multimodal ELBO
Thomas M. Sutter, Imant Daunhawer, Julia E Vogt
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 |
|---|---|---|---|
| Use the MoPoE posterior instead of the MVAE posterior | Multimodal VAE · MNIST-SVHN-Text (all modalities) | latent classification accuracy: 0.96 ±0.02 -> 0.98 ±0.01 | Improved |
| Use the MoPoE posterior instead of the MMVAE posterior | Multimodal VAE · MNIST-SVHN-Text (all modalities) | generation coherence: 0.28 -> 0.31 | Improved |
| Use the MoPoE posterior instead of the MVAE posterior | Multimodal VAE · MNIST-SVHN-Text (joint) | test log-likelihood: -1790 ±3.3 -> -1819 ±5.7 | Hurt |
| Use the MoPoE posterior instead of the MMVAE posterior | Multimodal VAE · CelebA (image and text; joint latent) | mean average precision: 0.35 -> 0.39 | Improved |
| Use the MoPoE posterior instead of the MMVAE posterior | Multimodal VAE · CelebA (image to text) | generation mean average precision: 0.14 -> 0.15 | Mixed |
| Use the MoPoE posterior instead of the MMVAE posterior | Multimodal VAE · CelebA (text to image) | generation mean average precision: 0.41 -> 0.43 | Improved |
| Use the MoPoE posterior instead of the MMVAE posterior | Multimodal VAE · PolyMNIST (joint generation) | joint coherence: 20.0 (±1.9) -> 12.1 (±1.6) percent | Hurt |
| Optimize posterior approximations for all modality subsets | MoPoE-VAE · PolyMNIST | conditional coherence | Improved |
| Combine subset PoEs with a mixture over subsets rather than using only the full-set PoE | Multimodal VAE · MNIST-SVHN-Text | generation coherence: 0.12 -> 0.31 | Improved |
Closest papers by meaning
- Multimodal Generative Learning Utilizing Jensen-Shannon-Divergence (NeurIPS 2020 · similarity 0.89)
- Learning Multimodal Latent Generative Models with Energy-Based Prior (ECCV 2024 · similarity 0.86)
- Aggregation of Dependent Expert Distributions in Multimodal Variational Autoencoders (ICML 2025 · similarity 0.85)
- Emu: Generative Pretraining in Multimodality (ICLR 2024 · similarity 0.83)
- Generative Multimodal Models are In-Context Learners (CVPR 2024 · similarity 0.83)
- On the Out-Of-Distribution Generalization of Large Multimodal Models (CVPR 2025 · similarity 0.83)
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.
