Research map / Reinforcement learning / Visual and representation RL
Avoid What You Know: Divergent Trajectory Balance for GFlowNets
Pedro Dall'Antonia, Tiago Silva, Daniel Csillag, Salem Lahlou, Diego Mesquita
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 |
|---|---|---|---|
| add the ACE exploration GFlowNet trained with DTB | canonical GFlowNet · Rings / 8 Gaussians (Lazy Random Walk) | TV distance | Improved |
| add the ACE exploration GFlowNet | epsilon-greedy, AT and SA GFlowNets · Grid world | convergence to target | Improved |
| add the ACE exploration GFlowNet | baseline GFlowNets · Bit sequences (K 32, 64) | top-K reward discovery | Improved |
| add the ACE exploration GFlowNet | baseline GFlowNets · Sequence design | discovery rate of high-reward states | Improved |
| add the ACE exploration GFlowNet | baseline GFlowNets · Quadratic Knapsack (K 128, 256) | sample efficiency | Improved |
| add the ACE exploration GFlowNet | baseline GFlowNets · AMP design | unique high-activity AMPs found | Improved |
| raise beta from 0.25 to 1 in DTB (ablation) | ACE · Rings | TV distance: beta 0.25 -> 1 | Hurt |
| raise alpha to 0.4 or more at beta=1 (ablation) | ACE · Rings | TV distance: alpha >= 0.4 at beta = 1 | Hurt |
| vary alpha over 0.1, 0.2, 0.4, 0.7 at beta=0.25 (ablation) | ACE · Rings | TV distance: alpha 0.1 to 0.7 | No clear effect |
| raise alpha at beta=0.5 (ablation) | ACE · Rings | convergence speed: beta = 0.5 | Hurt |
Closest papers by meaning
- Pre-Training and Fine-Tuning Generative Flow Networks (ICLR 2024 · similarity 0.92)
- Flow Factorization for Efficient Generative Flow Networks (AAAI 2025 · similarity 0.92)
- Symmetry-Aware GFlowNets (ICML 2025 · similarity 0.91)
- Looking Backward: Retrospective Backward Synthesis for Goal-Conditioned GFlowNets (ICLR 2025 · similarity 0.91)
- Pessimistic Backward Policy for GFlowNets (NeurIPS 2024 · similarity 0.91)
- Optimizing Backward Policies in GFlowNets via Trajectory Likelihood Maximization (ICLR 2025 · similarity 0.91)
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.
