Research map / Probabilistic and scientific modeling / Variational and generative sampling
Vector Quantized Wasserstein Auto-Encoder
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 |
|---|---|---|---|
| replace VQ-VAE quantization objective with Wasserstein alignment and learnable codeword marginals | VQ-WAE · CIFAR10, MNIST, SVHN, CelebA, FFHQ | rFID: 77.3 -> 54.3; 4.8 -> 2.33; 38.5 -> 23.4; 19.4 -> 12.2; 4.42 -> 4.20 | Helped |
| increase codebook size from 64 to 512 (ablation) | VQ-WAE · CIFAR10 | rFID: 73.5 -> 54.3 | Helped |
| increase codebook size from 64 to 512 (ablation) | VQ-WAE · MNIST | rFID: 5.6 -> 2.3 | Helped |
| learn πm instead of fixing peaked, Gaussian, or uniform initial marginals (ablation) | VQ-WAE · CIFAR10 | rFID: 63.77 -> 60.60; 68.87 -> 60.31; 56.06 -> 54.30 | Mixed |
| add KL regularization on πm with λr=1.0 rather than remove it (ablation) | VQ-WAE · CIFAR10 | rFID: 57.49 -> 54.30 | Helped |
| add KL regularization on πm (ablation) | VQ-WAE · CIFAR10 | training stability | Helped |
| change λ from 1e−2 to 1e−3 with λr=0.0 (ablation) | VQ-WAE · CIFAR10 | rFID: 68.99 -> 57.49 | Helped |
| change λ from 1e−3 to 1e−4 with λr=0.0 (ablation) | VQ-WAE · CIFAR10 | rFID: 57.49 -> 58.17 | Hurt |
| fix π to a uniform distribution rather than optimize it (ablation) | VQ-WAE · CIFAR10 | rFID | Hurt |
| use VQ-WAE codebook rather than SQ-VAE codebook for PixelCNN generation | VQ-WAE · CIFAR10, MNIST, SVHN, CelebA unconditional generation | FID: 103.78 -> 87.73; 8.93 -> 8.21; 31.26 -> 30.71; 29.5 -> 28.8 | Helped |
Closest papers by meaning
- Neural Discrete Representation Learning (NeurIPS 2017 · similarity 0.92)
- SQ-VAE: Variational Bayes on Discrete Representation with Self-annealed Stochastic Quantization (ICML 2022 · similarity 0.90)
- Finite Scalar Quantization: VQ-VAE Made Simple (ICLR 2024 · similarity 0.90)
- Restructuring Vector Quantization with the Rotation Trick (ICLR 2025 · similarity 0.89)
- Learning to Quantize for Training Vector-Quantized Networks (ICML 2025 · similarity 0.88)
- Information-theoretic Generalization Analysis for VQ-VAEs: A Role of Latent Variables (NeurIPS 2025 · similarity 0.87)
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.