Research map / Probabilistic and scientific modeling / Variational and generative sampling
Meta Flow Maps enable scalable reward alignment
Peter Potaptchik, Adhi Saravanan, Abbas Mammadov, Alvaro Prat, Michael Samuel Albergo, Yee Whye Teh
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 MFM one-step posterior samples instead of GLASS ODE rollouts for posterior sampling | MFM-XL/2 · ImageNet 256x256 posterior FID and value-function correlation | posterior FID / correlation | Improved |
| steer with MFM-G (N=1) instead of Best-of-N with N=1000 | MFM-XL/2 class tabby cat · ImageNet steering with ImageReward, PickScore, HPSv2 | average reward: over 100x fewer NFEs | Improved |
| use gradient-based MFM-G instead of gradient-free MFM-GF (ablation) | MFM steering · MNIST class-ratio matching | L2 to target class ratios | Improved |
| use MFM-Search instead of MFM-GF for gradient-free steering (ablation) | MFM steering · ImageNet tabby cat steering | reward | Improved |
| replace DPS with MFM-based steering | steering · GMM inverse problem and MNIST | distribution error / class-ratio L2 | Improved |
| increase number of Monte Carlo samples in MFM steering (ablation) | MFM-G / MFM-GF · GMM inverse problem | MMD and Sliced-Wasserstein | Improved |
| steer with HPSv2 or PickScore and evaluate on the other reward models | MFM-G · ImageNet steering, cross-reward evaluation | other reward scores | Improved |
| fine-tune with MFM-FT on HPSv2 | MFM-XL/2 · ImageNet all classes, HPSv2 prompt template | HPSv2, PickScore, ImageReward | Improved |
| increase K-step refinement from 1 to 4 steps (ablation) | MFM-XL/2 · ImageNet 256x256 | FID: 3.72 -> 1.97 | Improved |
Closest papers by meaning
- Align Your Flow: Scaling Continuous-Time Flow Map Distillation (NeurIPS 2025 · similarity 0.89)
- How to build a consistency model: Learning flow maps via self-distillation (NeurIPS 2025 · similarity 0.88)
- Test-time Scaling of Diffusions with Flow Maps (NeurIPS 2026 · similarity 0.88)
- PRECISE: SDE-Consistent Stochastic Sampling for RL Post-Training of Flow-Matching Models (NeurIPS 2026 · similarity 0.88)
- Aligning Flow Map Policies with Optimal $Q$-Guidance (NeurIPS 2026 · similarity 0.88)
- TMPO: Trajectory Matching Policy Optimization for Diverse and Efficient Diffusion Alignment (NeurIPS 2026 · 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.
