Research map / Probabilistic and scientific modeling / Variational and generative sampling
Unscented Autoencoder
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 sigma points instead of random posterior samples, and match the mean decoded image (ablation) | VAE · Fashion-MNIST | FID (sampling): 48.73 -> 30.39 | Helped |
| Use sigma points instead of random posterior samples, and match the mean decoded image (ablation) | VAE · CIFAR10 | FID (sampling): 120.6 -> 111.7 | Helped |
| Use sigma points instead of random posterior samples, and match the mean decoded image (ablation) | VAE · CelebA | FID (sampling): 68.53 -> 54.15 | Helped |
| Replace KL regularization with Wasserstein metric regularization (ablation) | VAE · Fashion-MNIST, CIFAR10, CelebA | FID (sampling): 48.73 -> 36.63; 120.6 -> 99.11; 68.53 -> 50.81 | Helped |
| Apply reconstruction loss to the mean decoded sigma-point image rather than each point's reconstruction (ablation) | UT-VAE₂ₓ · Fashion-MNIST, CIFAR10, CelebA | FID (sampling): 52.87 -> 40.30; 128.2 -> 115.4; 61.03 -> 57.42 | Helped |
| Add decoder regularization (ablation) | UT-VAE* · Fashion-MNIST | FID | Hurt |
| Add decoder regularization (ablation) | UT-VAE* · CIFAR10 and CelebA | FID | Helped |
| Increase sigma-point count from 1 to 8 (ablation) | UT-VAE with random sigma-point selection · Fashion-MNIST | FID (sampling): 52.10 -> 30.39 | Helped |
| Increase sigma-point count from 1 to 8 (ablation) | UT-VAE with random sigma-point selection · CIFAR10 | FID (sampling): 129.8 -> 110.8 | Helped |
| Increase sigma-point count from 1 to 8 (ablation) | UT-VAE with random sigma-point selection · CelebA | FID (sampling): 62.13 -> 54.15 | Helped |
| Use random pairs of sigma points instead of randomly selected sigma points (ablation) | UT-VAE · Fashion-MNIST and CelebA | FID | Hurt |
| Use largest-eigenvalue sigma-point pairs instead of random selection (ablation) | UT-VAE · CIFAR10 | FID | Helped |
Closest papers by meaning
- Variational Autoencoder with Implicit Optimal Priors (AAAI 2019 · similarity 0.92)
- From Variational to Deterministic Autoencoders (ICLR 2020 · similarity 0.92)
- MAE: Mutual Posterior-Divergence Regularization for Variational AutoEncoders (ICLR 2019 · similarity 0.91)
- NVAE: A Deep Hierarchical Variational Autoencoder (NeurIPS 2020 · similarity 0.89)
- A Contrastive Learning Approach for Training Variational Autoencoder Priors (NeurIPS 2021 · similarity 0.89)
- Explicitly Minimizing the Blur Error of Variational Autoencoders (ICLR 2023 · similarity 0.89)
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.