Research map / Probabilistic and scientific modeling / Variational and generative sampling
NVAE: A Deep Hierarchical Variational 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 |
|---|---|---|---|
| add normalizing flows in encoder (ablation) | NVAE · CIFAR-10 | bpd: 2.93 -> 2.91 | Helped |
| add normalizing flows in encoder (ablation) | NVAE · MNIST | NLL: 78.01 -> 78.19 | Hurt |
| replace WN with BN (ablation) | NVAE (ELU, L=40) · CIFAR-10 | bpd: 3.31 -> 3.22 | Helped |
| replace ELU with Swish (ablation) | NVAE (BN, L=40) · CIFAR-10 | bpd: 3.22 -> 3.16 | Helped |
| use depthwise separable cells in top-down generative model (ablation) | NVAE · CIFAR-10 | bpd: 3.11 -> 3.07 | Helped |
| use depthwise separable cells in bottom-up encoder (ablation) | NVAE · CIFAR-10 | bpd: 3.11 -> 3.12 | No clear effect |
| add residual Normal distribution parameterization (ablation) | NVAE · CIFAR-10 | test NLL bpd: 3.19 -> 3.16 | Helped |
| add spectral regularization (ablation) | NVAE · CIFAR-10 | test NLL: 3.18 -> 3.16 | Helped |
| add squeeze-and-excitation (ablation) | NVAE · CIFAR-10 | test NLL: 3.22 -> 3.16 | Helped |
| NVAE architecture vs prior non-autoregressive SOTA | VAE · CIFAR-10 | bpd: 2.98 -> 2.91 | Helped |
Closest papers by meaning
- Very Deep VAEs Generalize Autoregressive Models and Can Outperform Them on Images (ICLR 2021 · similarity 0.93)
- PixelVAE: A Latent Variable Model for Natural Images (ICLR 2017 · similarity 0.91)
- Unscented Autoencoder (ICML 2023 · similarity 0.89)
- MAE: Mutual Posterior-Divergence Regularization for Variational AutoEncoders (ICLR 2019 · similarity 0.89)
- Soft-IntroVAE: Analyzing and Improving the Introspective Variational Autoencoder (CVPR 2021 · similarity 0.89)
- Generating Diverse High-Fidelity Images with VQ-VAE-2 (NeurIPS 2019 · 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.