Research map / Probabilistic and scientific modeling / Variational and generative sampling
A Non-Negative Deep VAE: The Generalized Gamma Belief Network
Zhibin Duan, Tiansheng Wen, Muyao Wang, Bo Chen, Hao Zhang, Bo Chen, Fei Meng, Hongwei Liu, Mingyuan Zhou
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 |
|---|---|---|---|
| replace linear GBN decoder with non-linear generative network | GBN · 20NG | perplexity: 678 -> 589 | Improved |
| replace linear GBN decoder with non-linear generative network | GBN · R8 | perplexity: 657 -> 385 | Improved |
| use gamma latents with Weibull posterior in place of Gaussian latents | VDVAE · MNIST | NLL (nats): 78.07 -> 77.86 | Improved |
| use gamma latents with Weibull posterior in place of Gaussian latents | VDVAE · OMNIGLOT | NLL (nats): 86.93 -> 86.04 | Improved |
| use gamma latents with Weibull posterior in place of Gaussian latents | VDVAE · CIFAR-10 | bits/dim: 2.87 -> 2.84 | Improved |
| use gamma latents with Weibull posterior in place of Gaussian latents | VDVAE · CELEBA | bits/dim: 2.00 -> 1.95 | Improved |
| compare with Attention VAE (100 importance samples on CIFAR-10) | Generalized GBN · MNIST | NLL (nats): 77.63 vs 77.86 | Hurt |
| use sparse gamma latents instead of Gaussian latents without regularizer | VAE · 3D Shapes (disentanglement) | FactorVAE metric: 0.550 -> 0.841 | Improved |
| use sparse gamma latents instead of Gaussian latents without regularizer | VAE · 3D Cars (disentanglement) | FactorVAE metric: 0.753 -> 0.946 | Improved |
| use sparse gamma latents instead of Gaussian latents without regularizer | VAE · 2D Shapes (disentanglement) | FactorVAE metric: 0.617 -> 0.626 | Improved |
| compare gamma latents with BetaTCVAE regularized Gaussian | BetaTCVAE · 3D Shapes (disentanglement) | FactorVAE metric: 0.974 vs 0.841 | Hurt |
| compare gamma latents with BetaTCVAE regularized Gaussian | BetaTCVAE · 3D Cars (disentanglement) | FactorVAE metric: 0.929 -> 0.946 | Improved |
Closest papers by meaning
- Augmentable Gamma Belief Networks (JMLR 2016 · similarity 0.91)
- Sawtooth Factorial Topic Embeddings Guided Gamma Belief Network (ICML 2021 · similarity 0.89)
- Convolutional Poisson Gamma Belief Network (ICML 2019 · similarity 0.89)
- Deep Relational Topic Modeling via Graph Poisson Gamma Belief Network (NeurIPS 2020 · similarity 0.88)
- Multimodal Poisson Gamma Belief Network (AAAI 2018 · similarity 0.87)
- WHAI: Weibull Hybrid Autoencoding Inference for Deep Topic Modeling (ICLR 2018 · 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.
