Research map / Probabilistic and scientific modeling / Variational and generative sampling
Tilt Matching for Scalable Sampling and Fine-Tuning
Peter Potaptchik, Lee Cheuk Kit, Michael Samuel Albergo
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 |
|---|---|---|---|
| fine-tune with ITM (lambda=1) instead of Adjoint Matching (lambda=1) | Stable Diffusion 1.5 · ImageReward | ImageReward: 0.2170 -> 0.4465 | Improved |
| fine-tune with ITM (lambda=1) instead of Adjoint Matching (lambda=1) | Stable Diffusion 1.5 · HPSv2 | HPSv2: 0.2576 -> 0.2659 | Improved |
| fine-tune with ITM (lambda=1) instead of base SD 1.5 | Stable Diffusion 1.5 · ImageReward | ImageReward: 0.1873 -> 0.4465 | Improved |
| fine-tune with ITM (lambda=1) instead of base SD 1.5 | Stable Diffusion 1.5 · DreamSim | DreamSim (diversity): 0.3849 -> 0.3383 | Hurt |
| fine-tune with ITM (lambda=1) instead of base SD 1.5 | Stable Diffusion 1.5 · CLIPScore | CLIPScore: 0.2746 -> 0.2794 | Improved |
| use reward multiplier lambda=100 with Adjoint Matching instead of ITM lambda=1 | Stable Diffusion 1.5 · ImageReward | ImageReward: 0.4465 -> 0.7873 | Improved |
| use reward multiplier lambda=100 with Adjoint Matching instead of ITM lambda=1 | Stable Diffusion 1.5 · CLIPScore | CLIPScore: 0.2794 -> 0.2792 | No clear effect |
| use ITM instead of iDEM | LJ-13 · LJ-13 | ESS: 0.231 -> 0.507 | Improved |
| use ITM instead of DDS | LJ-13 · LJ-13 | energy W2: 24.61 -> 0.879 | Improved |
| use ITM instead of ASBS | LJ-55 · LJ-55 | energy W2: 28.10 -> 26.37 | Improved |
| use ITM instead of ASBS | LJ-55 · LJ-55 | Geo W2: 4.00 -> 3.98 | Improved |
| use low-variance ITM loss instead of weighted flow matching (ablation) | Tilt Matching · LJ-13 | gradient variance | Improved |
Closest papers by meaning
- Test-time Scaling of Diffusions with Flow Maps (NeurIPS 2026 · similarity 0.91)
- Elucidating Flow Matching ODE Dynamics via Data Geometry and Denoisers (ICML 2025 · similarity 0.89)
- PRECISE: SDE-Consistent Stochastic Sampling for RL Post-Training of Flow-Matching Models (NeurIPS 2026 · similarity 0.89)
- Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch (NeurIPS 2025 · similarity 0.89)
- Equilibrium Matching: Generative Modeling with Implicit Energy-Based Models (NeurIPS 2026 · similarity 0.88)
- Wasserstein Gradient Flows and Forward-Only Diffusion Are Not Enough for Multimodal Sampling (NeurIPS 2026 · similarity 0.88)
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.
