Research map / Probabilistic and scientific modeling / Variational and generative sampling
ByPE-VAE: Bayesian Pseudocoresets Exemplar VAE
Qingzhong Ai, LIRONG HE, SHIYU LIU, Zenglin Xu
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 Gaussian prior with ByPE pseudocoreset prior | VAE · Dynamic MNIST | test NLL: 24.41 ± 0.06 -> 23.61 ± 0.03 | Improved |
| replace Gaussian prior with ByPE pseudocoreset prior | VAE · CIFAR10 | test NLL: 72.21 ± 0.08 -> 71.91 ± 0.02 | Improved |
| ByPE versus VampPrior | HVAE · CIFAR10 | test NLL: 71.54 ± 0.04 -> 71.38 ± 0.01 | Improved |
| ByPE versus VampPrior | ConvHVAE · Dynamic MNIST | test NLL: 22.94 ± 0.02 -> 22.84 ± 0.02 | Improved |
| ByPE versus Exemplar prior | fully convolutional VAE · CelebA | test NLL: 185.03 ± 1.46 -> 182.11 ± 1.10 | Improved |
| ByPE pseudocoreset of 500 instead of 25000 exemplars | Exemplar VAE · Dynamic MNIST | training time (s/epoch): 35.45 -> 13.19 | Improved |
| ByPE pseudocoreset of 500 instead of 25000 exemplars | Exemplar VAE · CIFAR10 | training time (s/epoch): 66.85 -> 17.30 | Improved |
| ByPE versus Exemplar VAE with 25000 exemplars | ByPE-VAE · Fashion MNIST | test NLL: 20.81 vs 20.85 | No clear effect |
| update pseudocoreset every k epochs, k from 1 to 100 (ablation) | ByPE-VAE · MNIST | test NLL: 23.60 -> 23.70 | No clear effect |
| vary lambda mixing weight of real and augmented data from 0 to 0.5 (ablation) | ByPE-VAE · permutation invariant MNIST | test error (%): 1.42 ± 0.08 -> 0.83 ± 0.05 | Improved |
| raise lambda from 0.5 to 1.0 (ablation) | ByPE-VAE · permutation invariant MNIST | test error (%): 0.83 ± 0.05 -> 1.31 ± 0.00 | Hurt |
| sample latents from prior instead of posterior for augmentation (ablation) | ByPE-VAE · permutation invariant CIFAR10 | test error (%): 49.00 ± 0.04 -> 46.60 ± 0.13 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Exemplar VAE (25000) — Exemplar VAE: Linking Generative Models, Nearest Neighbor Retrieval, and Data Augmentation (NeurIPS 2020) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Exemplar VAE: Linking Generative Models, Nearest Neighbor Retrieval, and Data Augmentation (NeurIPS 2020 · similarity 0.86)
- NVAE: A Deep Hierarchical Variational Autoencoder (NeurIPS 2020 · similarity 0.86)
- Variational Autoencoder with Implicit Optimal Priors (AAAI 2019 · similarity 0.86)
- A Contrastive Learning Approach for Training Variational Autoencoder Priors (NeurIPS 2021 · similarity 0.85)
- Very Deep VAEs Generalize Autoregressive Models and Can Outperform Them on Images (ICLR 2021 · similarity 0.85)
- Bayesian Basis Function Approximation for Scalable Gaussian Process Priors in Deep Generative Models (ICML 2025 · similarity 0.84)
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.
