Research map / Probabilistic and scientific modeling / Variational and generative sampling
Hierarchical Semi-Implicit Variational Inference with Application to Diffusion Model Acceleration
Longlin Yu, Tianyu Xie, Yu Zhu, Tong Yang, Xiangyu Zhang, Cheng Zhang
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 |
|---|---|---|---|
| increase HSIVI-SM layers T from 1 to 5 (ablation) | HSIVI-SM · Conditioned diffusion (300-d) | Frobenius distance: 0.0886 -> 0.0333 | Improved |
| increase HSIVI-LB layers T from 1 to 5 (ablation) | HSIVI-LB · Conditioned diffusion (300-d) | Frobenius distance: 0.0883 -> 0.0433 | Improved |
| use score-matching objective instead of ELBO-based LB at T=5 | HSIVI · Conditioned diffusion (300-d) | Frobenius distance: 0.0433 -> 0.0333 | Improved |
| replace single-layer SIVI with 5-layer HSIVI | SIVI · 8-mode Gaussian mixture | mode coverage | Improved |
| replace single-layer SIVI-SM with 5-layer HSIVI-SM | SIVI-SM · Conditioned diffusion (300-d) | posterior variance fit | Improved |
| use non-isotropic instead of isotropic conditional layers (NFE=5) (ablation) | HSIVI-SM · CIFAR-10 | FID: 7.33 -> 6.27 | Improved |
| use non-isotropic instead of isotropic conditional layers (NFE=10) (ablation) | HSIVI-SM · CIFAR-10 | FID: 4.78 -> 4.31 | Improved |
| use non-isotropic instead of isotropic conditional layers (NFE=15) (ablation) | HSIVI-SM · CIFAR-10 | FID: 4.46 -> 4.17 | Improved |
| HSIVI-SM instead of DDIM at 5 steps | diffusion sampler · CIFAR-10 | FID: 41.53 -> 6.27 | Improved |
| HSIVI-SM instead of DDIM at 5 steps | diffusion sampler · ImageNet 64x64 | FID: 147.03 -> 40.43 | Improved |
| HSIVI-SM instead of DPM-Solver-fast at 15 NFE | diffusion sampler · CIFAR-10 | FID: 4.67 -> 4.17 | Improved |
| HSIVI-SM instead of DPM-Solver-fast at 10 NFE | diffusion sampler · CelebA | FID: 6.76 -> 3.09 | Improved |
Closest papers by meaning
- Kernel Semi-Implicit Variational Inference (ICML 2024 · similarity 0.92)
- Revisiting Unbiased Implicit Variational Inference (ICML 2025 · similarity 0.89)
- Continuous Semi-Implicit Models (ICML 2025 · similarity 0.89)
- Particle Semi-Implicit Variational Inference (NeurIPS 2024 · similarity 0.89)
- Denoising Diffusion Variational Inference: Diffusion Models as Expressive Variational Posteriors (AAAI 2025 · similarity 0.88)
- Model-Informed Flows for Bayesian Inference (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.
