Research map / Probabilistic and scientific modeling / Molecules and proteins
Routing by Reaching: Composition of Pre-trained GFlowNets for Multi-Objective Generation
Seokwon Yoon, Youngbin Choi, Seunghyuk Cho, Seungbeom Lee, MoonJeong Park, Dongwoo Kim
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 |
|---|---|---|---|
| weight forward policies by reaching probability instead of uniform ensemble mixing (ablation) | GFlowNet mixing · 2D grid scalarization (2 objectives) | L1 error: 0.117 -> 0.003 | Improved |
| weight forward policies by reaching probability instead of uniform ensemble mixing (ablation) | GFlowNet mixing · 2D grid logical: Circle2 contrast Circle1 | L1 error: 0.390 -> 0.070 | Improved |
| increase number of objectives from 2 to 5 | Ours · 2D grid scalarization | L1 error: 0.003 -> 0.003 | No clear effect |
| increase number of objectives from 2 to 5 | MOGFN · 2D grid scalarization | L1 error: 0.021 -> 0.048 | Hurt |
| increase number of objectives from 2 to 5 | HN-GFN · 2D grid scalarization | L1 error: 0.017 -> 0.035 | Hurt |
| compose pre-trained GFlowNets instead of classifier guidance | logical operators · 2D grid logical: Shubert harmonic mean Diagonal | L1 error: 0.142 -> 0.180 | Hurt |
| compose pre-trained GFlowNets instead of classifier guidance | logical operators · 2D grid logical: Circle1 harmonic mean Circle2 | L1 error: 0.397 -> 0.229 | Improved |
| compose pre-trained GFlowNets instead of classifier guidance | three-objective harmonic mean · Fragment logical SEH x SA x QED | Target-bin samples (%): 40 -> 66 | Improved |
| compose pre-trained GFlowNets instead of classifier guidance | logical operators · Fragment logical generation time | Generation time (ms): 1900.71 -> 29.62 | Improved |
| compose pre-trained GFlowNets instead of preference-conditioned MOGFN | scalarization · QM9 ALL | Top-10 reward: 0.688 -> 0.727 | Improved |
| compose pre-trained GFlowNets instead of HN-GFN | scalarization · Fragment ALL | Hypervolume: 0.651 -> 0.674 | Improved |
| compose pre-trained GFlowNets instead of preference-conditioned baselines | scalarization · Fragment SEH-SA | Top-10 reward: 0.839 -> 0.836 | No clear effect |
Closest papers by meaning
- Order-Preserving GFlowNets (ICLR 2024 · similarity 0.92)
- Pre-Training and Fine-Tuning Generative Flow Networks (ICLR 2024 · similarity 0.92)
- Pessimistic Backward Policy for GFlowNets (NeurIPS 2024 · similarity 0.90)
- Local Search GFlowNets (ICLR 2024 · similarity 0.90)
- Flow Factorization for Efficient Generative Flow Networks (AAAI 2025 · similarity 0.90)
- Looking Backward: Retrospective Backward Synthesis for Goal-Conditioned GFlowNets (ICLR 2025 · similarity 0.90)
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.
