Research map / Image generation / Text-to-image diffusion
Diffusion Alignment as Variataional Expectation-Maximization
Jaewoo Lee, Minsu Kim, Sanghyeok Choi, Inhyuck Song, Sujin Yun, Hyeongyu Kang, Woocheol Shin, Taeyoung Yun, Kiyoung Om, Jinkyoo Park
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 KL regularization to DAV (DAV-KL) (ablation) | DAV · Text-to-image | aesthetic score and ImageReward: Aesthetic 8.04 -> 6.99; ImageReward 0.95 -> 1.13; LPIPS-A 0.53 -> 0.58 | Mixed |
| amortize test-time-search posterior samples (DAV Posterior vs DAV) (ablation) | DAV · Text-to-image | aesthetic score: 8.04 -> 9.18 | Improved |
| DAV vs fine-tuning baselines DDPO and DRaFT | DAV · Text-to-image | aesthetic score: 6.83 / 7.22 -> 8.04 | Improved |
| DAV Posterior vs DAS test-time search | DAV Posterior · Text-to-image | aesthetic score: 7.22 -> 9.18 | Improved |
| DAV vs DAS on diversity | DAV Posterior · Text-to-image | LPIPS-A: 0.65 -> 0.53 | Hurt |
| replace E-step posterior search with search-and-distill (skip posterior update) (ablation) | DAV · Text-to-image | ELBO and aesthetic score | Hurt |
| replace test-time search with reward reweighting of trajectories (ablation) | DAV · Text-to-image | ELBO and aesthetic score | Hurt |
| lower KL weight alpha (ablation) | DAV · Text-to-image | reward vs over-optimization | Mixed |
| DAV on DNA enhancer design vs DRAKES, DDPO, VIDD | DAV · DNA enhancer sequence design | reward-diversity-naturalness tradeoff | Improved |
Closest papers by meaning
- Inference-Time Alignment of Diffusion Models with Direct Noise Optimization (ICML 2025 · similarity 0.92)
- Test-time Alignment of Diffusion Models without Reward Over-optimization (ICLR 2025 · similarity 0.91)
- Direct Distributional Optimization for Provable Alignment of Diffusion Models (ICLR 2025 · similarity 0.91)
- Towards Better Alignment: Training Diffusion Models with Reinforcement Learning Against Sparse Rewards (CVPR 2025 · similarity 0.90)
- Composition and Alignment of Diffusion Models using Constrained Learning (NeurIPS 2025 · similarity 0.89)
- Diffusion Tree Sampling: Scalable inference‑time alignment of diffusion models (NeurIPS 2025 · similarity 0.89)
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.
