Research map / Probabilistic and scientific modeling / Variational and generative sampling
Variational Autoencoder with Arbitrary Conditioning
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 VAEAC instead of the strongest listed imputation baseline | VAEAC · WhiteWine, Yeast, Mushroom, and Phishing | NRMSE/PFC: WhiteWine: 0.878 ± 0.009 -> 0.850 ± 0.007; Yeast: 1.01 ± 0.01 -> 0.94 ± 0.01; Mushroom: 0.249 ± 0.006 -> 0.244 ± 0.002; | Helped |
| Use VAEAC instead of MissForest on Zoo | VAEAC · Zoo | PFC: 0.16 ± 0.02 -> 0.16 ± 0.02 | No clear effect |
| Use VAEAC imputations for downstream prediction | VAEAC · WhiteWine, Yeast, Mushroom, Zoo, Phishing | R2-score/accuracy: WhiteWine: 0.17 ± 0.01 vs 0.17 ± 0.01; Yeast: 0.43 ± 0.01 vs 0.42 ± 0.02; Mushroom: 0.983 ± 0.002 vs 0.972 ± 0.003; Zoo: | Mixed |
| Generate 10 inpaintings rather than 1 and select the best PSNR | VAEAC · CelebA Center, Pattern, Half, O1 and O2 masks | PSNR: Center: 22.1 -> 23.7; Pattern: 21.4 -> 23.3; Half: 14.9 -> 17.4; O1: 20.8 -> 22.0; O2: 21.0 -> 22.2 | Helped |
| Use VAEAC instead of Context Encoder | VAEAC, 1 sample · CelebA Center, Pattern and Random masks | PSNR: Center: 21.3 -> 22.1; Pattern: 19.2 -> 21.4; Random: 20.6 -> 29.3 | Helped |
| Use VAEAC instead of Context Encoder on Half mask | VAEAC, 1 sample · CelebA Half mask | PSNR: 15.5 -> 14.9 | Hurt |
| Use VAEAC instead of GFC | VAEAC, 1 sample · CelebA O1 and O2 masks | PSNR: O1: 20.0 -> 20.8; O2: 19.8 -> 21.0 | Helped |
| Replace parallel conditional sampling with UM's autoregressive sampling | UM versus VAEAC · MNIST | negative log-likelihood: VAEAC 61; UM 41 | Mixed |
| Use VAEAC rather than UM for sample generation | VAEAC versus UM · MNIST | test time (100 samples generation): 1s -> 0.7ms | Helped |
| Add skip-connections between prior and generative networks (ablation) | VAEAC · Image inpainting | log-likelihood and image realism | Helped |
| Remove skip-connections (ablation) | VAEAC · Image inpainting | image realism | Hurt |
| Use GSNN/hybrid objective rather than the VAEAC variational objective (ablation) | VAEAC · Multimodal conditional distributions | mode fidelity | Hurt |
Closest papers by meaning
- From Variational to Deterministic Autoencoders (ICLR 2020 · similarity 0.88)
- Discrete Variational Autoencoders (ICLR 2017 · similarity 0.88)
- Shape your Space: A Gaussian Mixture Regularization Approach to Deterministic Autoencoders (NeurIPS 2021 · similarity 0.88)
- Laplacian Autoencoders for Learning Stochastic Representations (NeurIPS 2022 · similarity 0.88)
- Stick-Breaking Variational Autoencoders (ICLR 2017 · similarity 0.88)
- Variational Learning of Fractional Posteriors (ICML 2025 · similarity 0.86)
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.