Research map / Reinforcement learning / RL from feedback
Value Matching: Scalable and Gradient-Free Reward-Guided Flow Adaptation
Cristian Perez Jensen, Luca Schaufelberger, Riccardo De Santi, Kjell Jorner, Andreas Krause
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 |
|---|---|---|---|
| adapt FlowMol with Value Matching on QED reward (λ=500) | FlowMol · FlowMol GEOM-Drugs, QED reward | stable molecules (%): 49.5 -> 67.6 | Improved |
| adapt FlowMol with Value Matching on QED reward | FlowMol · FlowMol GEOM-Drugs, QED reward | QED: 0.42 -> 0.49 | Improved |
| adapt FlowMol with Value Matching on dipole moment reward | FlowMol · FlowMol GEOM-Drugs, dipole moment reward | dipole moment (Debye): 6.4 -> 7.5 | Improved |
| use VM instead of classifier guidance at moderate reward scale | VM · CIFAR-10 aesthetic reward | training stability / reward | Improved |
| use online VM instead of offline classifier guidance | VM · 2D toy environment | KL to tilted distribution | Improved |
| use VM instead of DDPO/DPOK fine-tuning | VM · CIFAR-10 (compression, incompression, aesthetic) | reward and diversity | Improved |
| use VM instead of CT-PPO | VM · CIFAR-10 rewards | reward | No clear effect |
| use VM instead of fine-tuning in memory | VM · SD2 | memory (GB): 250 -> 12 | Improved |
| increase value network size from 1.8M to 92.3M parameters (ablation) | VM · CIFAR-10 aesthetic reward (λ=100) | aesthetic reward: 4.00 -> 3.26 | No clear effect |
| use VM guidance instead of SVDD inference-time search | VM · CIFAR-10 aesthetic reward | sampling cost / reward | Improved |
| add VM value guidance at sampling | VM · SD2 sampling | time (s): 122.02 -> 127.44 | Hurt |
Closest papers by meaning
- Reward Score Matching: Unifying Reward-based Fine-tuning for Flow and Diffusion Models (NeurIPS 2026 · similarity 0.90)
- Value Gradient Guidance for Flow Matching Alignment (NeurIPS 2025 · similarity 0.89)
- Online Reward-Weighted Fine-Tuning of Flow Matching with Wasserstein Regularization (ICLR 2025 · similarity 0.89)
- Efficient Adjoint Matching for Fine-tuning Diffusion Models (NeurIPS 2026 · similarity 0.88)
- Flow Density Control: Generative Optimization Beyond Entropy-Regularized Fine-Tuning (NeurIPS 2025 · similarity 0.88)
- On the Guidance of Flow Matching (ICML 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.
