AI Paper Atlas

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

ICML 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Structure-based 3D molecule generation · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
Wrap fine-tuning in CFO augmented-Lagrangian loopAdjoint Matching · 2D Gaussian mixtureexpected constraint: 2.47 -> 0.12Improved
Wrap fine-tuning in CFOAdjoint Matching · 2D Gaussian mixtureexpected reward: −2.93 -> −4.75Hurt
Wrap fine-tuning in CFOAdjoint Matching · FlowMol dipole/energyexpected energy (Ha): −78.31 -> −82.28Improved
Wrap fine-tuning in CFOAdjoint Matching · FlowMol dipole/energydipole moment (D): 8.37 -> 8.39No clear effect
Use gradient-free DiffusionNFT as inner solver of CFODiffusionNFT · FlowMol dipole/energyexpected energy (Ha): −78.67 -> −80.72Improved
Use CFO instead of unconstrained fine-tuningAdjoint Matching · FlowMol GEOM Drugsvalidity (%): 34 -> 9 (PRE -> CFO); AM 4Hurt
Use CFO instead of unconstrained AMAdjoint Matching · FlowMol GEOM Drugsper-sample feasibility: 40.6% -> 61.4%Improved
Adapt penalty online (CFO) instead of manually tuned fixed muFixed-mu Lagrangian · FlowMol dipole/energyconstraint satisfaction and rewardImproved
Small fixed penalty mu<=0.01 (ablation)Fixed-mu Lagrangian · FlowMol dipole/energyconstraint satisfaction: 8.34 D, −78.94 HaHurt
Large fixed penalty mu=50.0 (ablation)Fixed-mu Lagrangian · FlowMol dipole/energydipole moment (D): 6.69Hurt
Increase outer iterations K (3 -> 100) under fixed budget (ablation)CFO · 2D Gaussian mixtureconstraint violation vs reward: 0.40 -> 0.10 violation; reward −5.91 at K=100Mixed
Intermediate K=20 (ablation)CFO · 2D Gaussian mixtureviolation and reward: 0.12 violation, −4.75 rewardImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy