Research map / Probabilistic and scientific modeling / Variational and generative sampling
Negative Binomial Variational Autoencoders for Overdispersed Latent Modeling
Yixuan Zhang, Jinhao Sheng, Wenxin Zhang, Quyu Kong, Feng 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 Poisson latents with negative binomial latents (MC-G variant) | P-VAE · MNIST | FID@5k: 105.3678 -> 79.6727 | Improved |
| replace Poisson latents with negative binomial latents (MC-G variant) | P-VAE · CIFAR 16x16 | FID@5k: 60.3653 -> 40.2788 | Improved |
| replace Poisson latents with negative binomial latents (DS-C variant) | P-VAE · CIFAR 16x16 | MSE: 0.0357 -> 0.0189 | Improved |
| replace Poisson latents with negative binomial latents | P-VAE · CelebA-64 | SSIM: 0.6354 -> 0.6329 | Hurt |
| use continuous-time reparameterization instead of Gumbel-Softmax with dispersion-sharing KL (ablation) | NegBio-VAE · CIFAR 16x16 | SSIM: 0.6328 -> 0.8089 | Improved |
| use Gumbel-Softmax instead of continuous-time reparameterization with Monte Carlo KL (ablation) | NegBio-VAE · CIFAR 16x16 | FID@5k: 67.2898 -> 40.2788 | Improved |
| use NegBio-VAE latents for MNIST fragmentation prediction vs P-VAE | NegBio-VAE DS-C · MNIST fragmentation prediction (N=200, latent 100) | accuracy: 0.736 -> 0.811 | Improved |
| use NegBio-VAE latents for few-shot logistic regression vs best baseline | NegBio-VAE · MNIST (20-shot) | accuracy: 0.838 -> 0.865 | Improved |
| use NegBio-VAE latents for few-shot kNN vs C-VAE | NegBio-VAE · MNIST (20-shot kNN) | accuracy: 0.791 -> 0.786 | Hurt |
| increase number of Monte Carlo samples from 5 to 25 (ablation) | NegBio-VAE MC variants · MNIST | MSE, SSIM, FID, KID | No clear effect |
| increase beta scaling from small (0.2-0.4) to about 1.0 (ablation) | NegBio-VAE · MNIST | reconstruction vs FID | Mixed |
| use stronger decoder (MLP or convolutional) instead of linear with linear encoder (ablation) | NegBio-VAE · MNIST | MSE, SSIM, FID, KID | Improved |
Closest papers by meaning
- Poisson Variational Autoencoder (NeurIPS 2024 · similarity 0.89)
- Learning identifiable and interpretable latent models of high-dimensional neural activity using pi-VAE (NeurIPS 2020 · similarity 0.87)
- Multi-modal Gaussian Process Variational Autoencoders for Neural and Behavioral Data (ICLR 2024 · similarity 0.86)
- Brain-like Variational Inference (NeurIPS 2025 · similarity 0.86)
- Multi-Modal Latent Variables for Cross-Individual Primary Visual Cortex Modeling and Analysis (AAAI 2025 · similarity 0.86)
- Probabilistic Decomposed Linear Dynamical Systems for Robust Discovery of Latent Neural Dynamics (NeurIPS 2024 · similarity 0.85)
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.
