Research map / Probabilistic and scientific modeling / Molecules and proteins
Constrained Flow Optimization via Sequential Fine-Tuning for Molecular Design
Sven Gutjahr, Riccardo De Santi, Luca Schaufelberger, 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 |
|---|---|---|---|
| Wrap fine-tuning in CFO augmented-Lagrangian loop | Adjoint Matching · 2D Gaussian mixture | expected constraint: 2.47 -> 0.12 | Improved |
| Wrap fine-tuning in CFO | Adjoint Matching · 2D Gaussian mixture | expected reward: −2.93 -> −4.75 | Hurt |
| Wrap fine-tuning in CFO | Adjoint Matching · FlowMol dipole/energy | expected energy (Ha): −78.31 -> −82.28 | Improved |
| Wrap fine-tuning in CFO | Adjoint Matching · FlowMol dipole/energy | dipole moment (D): 8.37 -> 8.39 | No clear effect |
| Use gradient-free DiffusionNFT as inner solver of CFO | DiffusionNFT · FlowMol dipole/energy | expected energy (Ha): −78.67 -> −80.72 | Improved |
| Use CFO instead of unconstrained fine-tuning | Adjoint Matching · FlowMol GEOM Drugs | validity (%): 34 -> 9 (PRE -> CFO); AM 4 | Hurt |
| Use CFO instead of unconstrained AM | Adjoint Matching · FlowMol GEOM Drugs | per-sample feasibility: 40.6% -> 61.4% | Improved |
| Adapt penalty online (CFO) instead of manually tuned fixed mu | Fixed-mu Lagrangian · FlowMol dipole/energy | constraint satisfaction and reward | Improved |
| Small fixed penalty mu<=0.01 (ablation) | Fixed-mu Lagrangian · FlowMol dipole/energy | constraint satisfaction: 8.34 D, −78.94 Ha | Hurt |
| Large fixed penalty mu=50.0 (ablation) | Fixed-mu Lagrangian · FlowMol dipole/energy | dipole moment (D): 6.69 | Hurt |
| Increase outer iterations K (3 -> 100) under fixed budget (ablation) | CFO · 2D Gaussian mixture | constraint violation vs reward: 0.40 -> 0.10 violation; reward −5.91 at K=100 | Mixed |
| Intermediate K=20 (ablation) | CFO · 2D Gaussian mixture | violation and reward: 0.12 violation, −4.75 reward | Improved |
Closest papers by meaning
- Flow Density Control: Generative Optimization Beyond Entropy-Regularized Fine-Tuning (NeurIPS 2025 · similarity 0.91)
- On the Design Space of Discrete Diffusion Online Adaptation for Molecular Optimization (NeurIPS 2026 · similarity 0.89)
- Bridging Model-Based Optimization and Generative Modeling via Conservative Fine-Tuning of Diffusion Models (NeurIPS 2024 · similarity 0.88)
- SynFlowNet: Design of Diverse and Novel Molecules with Synthesis Constraints (ICLR 2025 · similarity 0.88)
- Controllable 3D Molecular Generation for Structure-Based Drug Design Through Bayesian Flow Networks and Gradient Integration (NeurIPS 2025 · similarity 0.88)
- Uncertainty-Aware Multi-Objective Reinforcement Learning-Guided Diffusion Models for 3D De Novo Molecular Design (NeurIPS 2025 · 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.
