Research map / Probabilistic and scientific modeling / Variational and generative sampling
On the Limitations of Multimodal VAEs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 modality sub-sampling in place of a unimodal VAE | MMVAE and MoPoE-VAE · PolyMNIST, Translated-PolyMNIST, and CUB images | FID | Hurt |
| Remove modality sub-sampling | MVAE · PolyMNIST, Translated-PolyMNIST, and CUB images | FID | Helped |
| Add modalities (ablation) | MMVAE and MoPoE-VAE · PolyMNIST and Translated-PolyMNIST | FID | Hurt |
| Add modalities (ablation) | MVAE · Translated-PolyMNIST | FID | Hurt |
| Use MMVAE or MoPoE-VAE instead of MVAE | Multimodal VAEs · PolyMNIST | leave-one-out coherence | Helped |
| Increase beta (ablation) | MVAE · PolyMNIST | leave-one-out coherence | Helped |
| Use Translated-PolyMNIST, where shared information cannot be predicted in expectation across modalities | MMVAE, MoPoE-VAE, and MVAE · Translated-PolyMNIST | generative coherence | Hurt |
| Conditionally generate images from captions on real-image CUB | MMVAE, MoPoE-VAE, and MVAE · CUB | generative coherence | Hurt |
Closest papers by meaning
- MMVAE+: Enhancing the Generative Quality of Multimodal VAEs without Compromises (ICLR 2023 · similarity 0.94)
- Aggregation of Dependent Expert Distributions in Multimodal Variational Autoencoders (ICML 2025 · similarity 0.92)
- Multimodal Generative Models for Scalable Weakly-Supervised Learning (NeurIPS 2018 · similarity 0.92)
- Hölder++: Improving Quality-Coherence Trade-off in Multimodal VAEs (ICML 2026 · similarity 0.91)
- Deep Generative Clustering with Multimodal Diffusion Variational Autoencoders (ICLR 2024 · similarity 0.91)
- Unity by Diversity: Improved Representation Learning for Multimodal VAEs (NeurIPS 2024 · similarity 0.91)
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.