Research map / Reinforcement learning / Visual and representation RL
PRO-MOF: Policy Optimization with Universal Atomistic Models for Controllable MOF Generation
Zicheng Liu, Ben Fei, Di Huang
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 Pass@K (K=3) reward instead of Pass@1 in GRPO (ablation) | PRO-MOF · Target PLD 6.0 A | success rate: 3.1 -> 7.8 | Improved |
| use Pass@K instead of Pass@1 (ablation) | PRO-MOF · Target PLD 6.0 A | diversity: 0.31 -> 0.65 | Improved |
| use UMA reward instead of simpler proxy (ablation) | PRO-MOF · Target PLD 6.0 A | success rate: 1.9 -> 7.8 | Improved |
| use UMA reward instead of simpler proxy (ablation) | PRO-MOF · Target PLD 6.0 A | diversity: 0.45 -> 0.65 | Improved |
| use Pass@K instead of Pass@1 (ablation) | PRO-MOF · Max CO2 working capacity | success rate: 8.1 -> 10.3 | Improved |
| use Pass@K instead of Pass@1 (ablation) | PRO-MOF · Min formation energy | success rate: 7.2 -> 12.4 | Improved |
| RL fine-tuning with UMA reward on MOFFlow-2 | MOFFlow-2 · Max CO2 working capacity | success rate: 4.0 -> 10.3 | Improved |
| RL fine-tuning versus genetic algorithm with UMA fitness | PRO-MOF · Min formation energy | success rate: 5.5 -> 12.4 | Improved |
| Pass@1 reward in RL (ablation) | PRO-MOF · training dynamics | structural diversity | Hurt |
Closest papers by meaning
- Flexible MOF Generation with Torsion-Aware Flow Matching (NeurIPS 2025 · similarity 0.89)
- MOFDiff: Coarse-grained Diffusion for Metal-Organic Framework Design (ICLR 2024 · similarity 0.88)
- MOFFlow: Flow Matching for Structure Prediction of Metal-Organic Frameworks (ICLR 2025 · similarity 0.87)
- MACS: Multi-Agent Reinforcement Learning for Optimization of Crystal Structures (NeurIPS 2025 · similarity 0.86)
- Co-PiLOT: Constrained Physics-Informed Latent Optimization for Target-Driven Inverse Design (NeurIPS 2026 · similarity 0.86)
- MANDREL: Modular Reinforcement Learning Pipelines for Material Discovery (AAAI 2024 · similarity 0.86)
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.
