Research map / Probabilistic and scientific modeling / Variational and generative sampling
On the Generative Utility of Cyclic Conditionals
Chang Liu, Haoyue Tang, Tao Qin, Jintao Wang, Tie-Yan Liu
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 |
|---|---|---|---|
| add compatibility loss when training (ablation) | CyGen-PT (SGLD) · Synthetic data, iteration 30000 | compatibility loss: 1.2\e{8} -> 4.6\e{3} | Improved |
| generate data with SGLD instead of Gibbs sampling (ablation) | CyGen-PT · Synthetic data | generation quality (qualitative) | Improved |
| pretrain likelihood model as a VAE (CyGen-PT vs CyGen) (ablation) | CyGen · Synthetic data | generation quality / posterior centering | No clear effect |
| smaller compatibility loss during training | CyGen · Synthetic data | generation quality | Improved |
| replace DAE with CyGen-PT | downstream classifier on latent · MNIST | accuracy (%): 98.0 -> 98.3 | Improved |
| replace VAE with CyGen-PT | downstream classifier on latent · MNIST | accuracy (%): 94.5 -> 98.3 | Improved |
| replace DAE with CyGen-PT | downstream classifier on latent · SVHN | accuracy (%): 74.5 -> 75.8 | Improved |
| replace VAE with CyGen-PT | downstream classifier on latent · SVHN | accuracy (%): 30.8 -> 75.8 | Improved |
| CyGen-PT vs GibbsNet (reported, different architecture) | downstream classifier on latent · SVHN | accuracy (%): 79.6 -> 75.8 | Hurt |
| replace DAE or VAE with CyGen-PT | generative model · SVHN | FID: 157 -> 102 | Improved |
| replace VAE with CyGen-PT | generative model · SVHN | FID: 128 -> 102 | Improved |
| use BiGAN/GibbsNet with the same probabilistic flow architecture | baseline generative models · MNIST and SVHN | generation quality | Hurt |
Closest papers by meaning
- A Likelihood Based Approach to Distribution Regression Using Conditional Deep Generative Models (ICML 2025 · similarity 0.88)
- A Theory for Conditional Generative Modeling on Multiple Data Sources (ICML 2025 · similarity 0.86)
- CoInD: Enabling Logical Compositions in Diffusion Models (ICLR 2025 · similarity 0.85)
- A Black-Box Debiasing Framework for Conditional Sampling (NeurIPS 2025 · similarity 0.85)
- Multimarginal Generative Modeling with Stochastic Interpolants (ICLR 2024 · similarity 0.84)
- Discrete Distribution Networks (ICLR 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.
