Research map / Probabilistic and scientific modeling / Variational and generative sampling
Multi-Source Neural Variational Inference
Richard Kurle, Stephan Günnemann, Patrick van der Smagt
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 |
|---|---|---|---|
| integrate redundant noisy sources using the hybrid elboac objective | MSNVI · MNIST-NO | negative ELBO: 94.95 -> 92.21 | Improved |
| use K=1 instead of K=16 (ablation) | MSNVI elboac · MNIST-TB | negative ELBO: 90.07 -> 90.81 | Hurt |
| use K=1 instead of K=16 (ablation) | MSNVI elboac · MNIST-QU | negative ELBO: 89.20 -> 90.17 | Hurt |
| use K=1 instead of K=16 (ablation) | MSNVI elboac · MNIST-NO | negative ELBO: 92.21 -> 94.49 | Hurt |
| learn individual source beliefs, then integrate them with a product of experts | MSNVI · MNIST-TB | negative ELBO (elboa evaluation): 102.20 -> 94.38 | Improved |
| learn individual source beliefs, then integrate them with a product of experts | MSNVI · MNIST-QU | negative ELBO (elboa evaluation): 120.46 -> 108.07 | Improved |
| learn the generative model for segmentation only while retaining all inference sources (ablation) | MSNVI elboa · Caltech-UCSD Birds 200 | segmentation accuracy: 0.808 -> 0.870 | Improved |
| learn the generative model for segmentation only while retaining all inference sources (ablation) | MSNVI elboac · Caltech-UCSD Birds 200 | segmentation accuracy: 0.810 -> 0.872 | Improved |
| learn generative models for both images and segmentations rather than segmentation only (ablation) | MSNVI elboa · Caltech-UCSD Birds 200 | segmentation accuracy: 0.870 -> 0.808 | Hurt |
| learn generative models for both images and segmentations rather than segmentation only (ablation) | MSNVI elboac · Caltech-UCSD Birds 200 | segmentation accuracy: 0.872 -> 0.810 | Hurt |
| integrate beliefs from redundant sensors | MSNVI pendulum inference · Pendulum | predictive uncertainty | Improved |
| detect a failed sensor using belief conflict and exclude it from integration | MSNVI pendulum inference · Pendulum | robustness to sensor failure | Improved |
Closest papers by meaning
- InfoVAE: Balancing Learning and Inference in Variational Autoencoders (AAAI 2019 · similarity 0.88)
- Deep Attentive Variational Inference (ICLR 2022 · similarity 0.87)
- Variational Autoencoders with Jointly Optimized Latent Dependency Structure (ICLR 2019 · similarity 0.87)
- Recursive Inference for Variational Autoencoders (NeurIPS 2020 · similarity 0.87)
- Variational Learning of Fractional Posteriors (ICML 2025 · similarity 0.87)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.85)
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.
