Research map / Image generation / Fast diffusion sampling
Efficient Tail-Aware Generative Optimization via Flow Model Fine-Tuning
Zifan Wang, Riccardo De Santi, Xiaoyu Mo, Michael M. Zavlanos, Andreas Krause, Karl Henrik Johansson
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 |
|---|---|---|---|
| Use left-CVaR pseudo-reward (L-TFFT) instead of expected-reward fine-tuning | Stable Diffusion v1-5 · ImageReward | L-CVaR0.2[r]: -0.370 -> -0.306 | Improved |
| Use L-TFFT instead of EXP-FT | Stable Diffusion v1-5 · ImageReward | E[r]: 0.852 -> 0.850 | No clear effect |
| Use L-TFFT instead of FDC under fixed 82 GPU hours | Stable Diffusion v1-5 · ImageReward | E[r]: 0.812 -> 0.850 | Improved |
| Use L-TFFT instead of EXP-FT | Stable Diffusion v1-5 · DreamSim diversity | DreamSim Var: 0.306 -> 0.289 | Hurt |
| Use L-TFFT instead of EXP-FT | Stable Diffusion v1-5 · HPSv2 | HPS: 0.278 -> 0.280 | Improved |
| Use right-CVaR (R-TFFT) instead of EXP-FT | FlowMol · xTB energy reward | R-CVaR0.9[r]: 167.1 -> 183.4 | Improved |
| Use R-TFFT instead of FDC (K=3) | FlowMol · xTB energy reward | R-CVaR0.9[r]: 172.5 -> 183.4 | Improved |
| Use R-TFFT instead of EXP-FT | FlowMol · xTB energy reward | E[r]: 29.8 -> 27.1 | Hurt |
| Use R-TFFT instead of EXP-FT | FlowMol · xTB energy reward | Validity: 78.3%±1.01% -> 85.7%±0.59% | Improved |
| Use R-TFFT instead of EXP-FT | FlowMol · xTB energy reward | SA score: 7.66 -> 7.70 | No clear effect |
| Solve CVaR dual threshold once offline instead of iterative FDC | TFFT · training cost | fine-tuning calls: K -> 1 | Improved |
Closest papers by meaning
- Flow Density Control: Generative Optimization Beyond Entropy-Regularized Fine-Tuning (NeurIPS 2025 · similarity 0.91)
- Bridging Model-Based Optimization and Generative Modeling via Conservative Fine-Tuning of Diffusion Models (NeurIPS 2024 · similarity 0.87)
- WTF?! Simulation-Free Reinforcement Learning with Wasserstein-Tilted Flow Maps (NeurIPS 2026 · similarity 0.87)
- Adjoint Matching: Fine-tuning Flow and Diffusion Generative Models with Memoryless Stochastic Optimal Control (ICLR 2025 · similarity 0.87)
- Reward Score Matching: Unifying Reward-based Fine-tuning for Flow and Diffusion Models (NeurIPS 2026 · similarity 0.87)
- Online Reward-Weighted Fine-Tuning of Flow Matching with Wasserstein Regularization (ICLR 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.
