Molecular Property Optimization on low tox proxy
1.14Normalized RewardRL
Evaluation Results
| Method | Links | |
|---|---|---|
| RLRL training budget=20002026.04 | 1.14 | |
| RLRL training budget=10002026.04 | 0.96 | |
| Reward Weighted Classifier-Free Guidance (RCFG)Inference-time sampling budget |YS|=642026.04 | 0.86 | |
| Reward Weighted Classifier-Free Guidance (RCFG)Inference-time sampling budget |YS|=322026.04 | 0.82 | |
| Reward Weighted Classifier-Free Guidance (RCFG)Inference-time sampling budget |YS|=162026.04 | 0.77 | |
| Reward Weighted Classifier-Free Guidance (RCFG)Inference-time sampling budget |YS|=82026.04 | 0.74 | |
| RLRL training budget=5002026.04 | 0.63 | |
| Reward Weighted Classifier-Free Guidance (RCFG)Inference-time sampling budget |YS|=42026.04 | 0.55 | |
| π(·|y*)2026.04 | 0.54 | |
| Reward Weighted Classifier-Free Guidance (RCFG)Inference-time sampling budget |YS|=22026.04 | 0.2 |