Research map / Learning theory and algorithms / Combinatorial and search optimization
Multi-Objective GFlowNets
Moksh Jain, Sharath Chandra Raparthy, Alex Hernández-García, Jarrid Rector-Brooks, Yoshua Bengio, Santiago Miret, Emmanuel Bengio
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 preference-conditional GFlowNet sampling instead of MOReinforce | 3 Bigrams · 3 Bigrams | HV: 0.015 -> 0.220 | Improved |
| Use preference-conditional GFlowNet sampling instead of MOReinforce | QM9 · QM9 | Diversity: 0.53 ± 0.08 -> 0.93 ± 0.00 | Improved |
| Use MOGFN-PC instead of MOReinforce | DNA sequence generation · DNA sequence generation | HV: 0.629 ± 0.002 -> 0.517 ± 0.006 | Hurt |
| Use MOGFN-PC instead of MOReinforce | DNA sequence generation · DNA sequence generation | Diversity: 0.6178 ± 0.209 -> 18.131 ± 0.981 | Improved |
| Use uniform preference sampling, Dirichlet(α=1), instead of skewed Dirichlet preferences (ablation) | MOGFN-PC · 3 Bigrams and fragment-based molecule generation | Normalized performance | Improved |
| Use weighted-sum scalarization instead of weighted-log-sum or weighted Tchebycheff (ablation) | MOGFN-PC · 3 Bigrams and fragment-based molecule generation | Normalized performance | Improved |
| Increase reward exponent β (ablation) | MOGFN-PC · Tasks in the β analysis | Pareto performance | Mixed |
| Increase reward exponent β (ablation) | MOGFN-PC · Tasks in the β analysis | Diversity | Hurt |
| Increase policy model size (ablation) | MOGFN-PC · 3 Bigrams | HV: 0.22 -> 0.255 -> 0.262 -> 0.270 | Improved |
| Increase policy model size (ablation) | MOGFN-PC · 3 Bigrams | Diversity: 19.79 -> 17.13 -> 17.53 -> 16.12 | Hurt |
| Use larger models rather than smaller models (ablation) | MOGFN-PC · 3 Bigrams | Reward: 0.44 -> 0.47 -> 0.49 -> 0.51 | Improved |
| Use preference thermometer encoding instead of raw preference values (ablation) | MOGFN-PC · QM9 | Performance | No clear effect |
Closest papers by meaning
- Order-Preserving GFlowNets (ICLR 2024 · similarity 0.91)
- Local Search GFlowNets (ICLR 2024 · similarity 0.87)
- Generative Flow Networks for Lead Optimization in Drug Design (Student Abstract) (AAAI 2025 · similarity 0.87)
- Learning to Build Solutions in Stochastic Matching Problems Using Flows (Student Abstract) (AAAI 2024 · similarity 0.86)
- ParetoFlow: Guided Flows in Multi-Objective Optimization (ICLR 2025 · similarity 0.86)
- Pre-Training and Fine-Tuning Generative Flow Networks (ICLR 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.
