Research map / Probabilistic and scientific modeling / Variational and generative sampling
Remedying Uncertainty Representations in Visual Inference Through Explaining-Away Variational Autoencoders
Josefina Catoni, Domonkos Martos, Ferenc Csikor, Enzo Ferrante, Diego Humberto Milone, Balazs Meszena, Gergő Orbán, Rodrigo Echeveste
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 |
|---|---|---|---|
| add a global multiplicative scaling latent s (EA-VAE) | standard VAE · natural image patches | posterior noise variance vs contrast | Improved |
| add scaling latent s | standard VAE · cMNIST and cFashionMNIST | signal mean saturation with contrast | Improved |
| add scaling latent s | standard VAE · MNIST morphing, 45 digit pairs | central uncertainty maximum: 31% | Improved |
| replace single-sample ELBO with multi-sample IWAE objective (ablation) | standard VAE · natural image patches | contrast-dependence of posterior variance | No clear effect |
| add MC Dropout at inference (ablation) | standard VAE · morphing and OOD experiments | uncertainty estimates | No clear effect |
| add scaling latent s | standard VAE · Chest X-ray blur and noise corruption | latent uncertainty | Improved |
| add scaling latent s | standard VAE · MNIST/ChestMNIST OOD, downstream MLP | predictive entropy: p < 10^-15 | Improved |
| use lognormal or gamma posterior for s instead of softplus-Laplace (ablation) | EA-VAE · natural image patches | qualitative uncertainty behaviour | No clear effect |
| add scaling latent s | standard VAE · training time on natural image patches | hours: 62.2 -> 67 | Hurt |
Closest papers by meaning
- Demystifying Inductive Biases for (Beta-)VAE Based Architectures (ICML 2021 · similarity 0.90)
- Consistency Regularization for Variational Auto-Encoders (NeurIPS 2021 · similarity 0.89)
- On Incorporating Inductive Biases into VAEs (ICLR 2022 · similarity 0.89)
- Towards Visually Explaining Variational Autoencoders (CVPR 2020 · similarity 0.89)
- Poisson Variational Autoencoder (NeurIPS 2024 · similarity 0.88)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.88)
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.
