Research map / Image generation / GANs
On the Quantitative Analysis of Decoder-Based Generative Models
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 |
|---|---|---|---|
| evaluate with AIS instead of KDE | log-likelihood estimator · Continuous MNIST GMMN-10 | log-likelihood accuracy | Helped |
| use AIS instead of IWAE bound | log-likelihood estimator · Binarized MNIST IWAE | log-likelihood (nats): -86.902 -> -85.679 | Helped |
| use AIS with 10x more intermediate distributions / IWAE with 10x more samples (ablation) | IWAE bound and AIS · Binarized MNIST | log-likelihood (nats): -85.679 -> -85.619 | No clear effect |
| initialize AIS with encoder q(z|x) (AIS+encoder) (ablation) | AIS on IWAE · Binarized MNIST | number of intermediate distributions: -85.754 -> -85.679 | Helped |
| use AIS instead of IWAE bound on continuous MNIST | VAE-50 · Continuous MNIST test | log-likelihood (nats): 826.325 -> 991.435 | Helped |
| use AIS instead of KDE | VAE-50 · Continuous MNIST test | log-likelihood (nats): 351.213 -> 991.435 | Helped |
| increase decoder from 10 to 50 latent dims | GMMN · Continuous MNIST test | log-likelihood (nats): 346.679 -> 593.472 | Helped |
| increase decoder from 10 to 50 latent dims | GAN · Continuous MNIST test | log-likelihood (nats): 328.772 -> 627.297 | Helped |
| increase decoder from 10 to 50 latent dims | VAE · Continuous MNIST test | log-likelihood (nats): 705.375 -> 991.435 | Helped |
| train with VAE objective instead of GAN | 50-dim decoder · Continuous MNIST test | log-likelihood (nats): 627.297 -> 991.435 | Helped |
| tune observation sigma per example instead of fixed (ablation) | GAN-50 · Continuous MNIST valid | log-likelihood (nats): 623.492 -> 702.699 | Helped |
| tune observation sigma per example instead of fixed (ablation) | GMMN-10 · Continuous MNIST valid | log-likelihood (nats): 332.360 -> 371.325 | Helped |
Closest papers by meaning
- Complexity Matters: Rethinking the Latent Space for Generative Modeling (NeurIPS 2023 · similarity 0.88)
- Adaptive Density Estimation for Generative Models (NeurIPS 2019 · similarity 0.87)
- Bias Correction of Learned Generative Models using Likelihood-Free Importance Weighting (NeurIPS 2019 · similarity 0.87)
- Unified Generation, Reconstruction, and Representation: Generalized Diffusion with Adaptive Latent Encoding-Decoding (ICML 2024 · similarity 0.86)
- Statistically Optimal Generative Modeling with Maximum Deviation from the Empirical Distribution (ICML 2024 · similarity 0.85)
- Improving Training Efficiency of Diffusion Models via Multi-Stage Framework and Tailored Multi-Decoder Architecture (CVPR 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.