Research map / Probabilistic and scientific modeling / Molecules and proteins
Loss-Guided Auxiliary Agents for Overcoming Mode Collapse in GFlowNets
Idriss Malek, Aya Laajil, Abhijith Sharma, Eric Moulines, Salem Lahlou
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 |
|---|---|---|---|
| train main GFlowNet with loss-guided auxiliary GFlowNet instead of on-policy only | GFlowNet TB · Hypergrid 2D 32x32 (sparse) | final L1 distance: 1460.00 -> 53.3 (x1e-6) | Improved |
| replace SAGFN novelty intrinsic reward with loss-guided auxiliary reward | GFlowNet TB · Hypergrid (2D and 4D, various sizes) | final L1 distance: 128x128: 7.07 -> 8.30 (x1e-6, SAGFN vs LGGFN) | No clear effect |
| use LGGFN simple additive loss reward instead of Adaptive Teachers multiplicative formulation | GFlowNet TB · Hypergrid 2D 32x32 | final L1 distance: 7.20 -> 5.33 (x1e-5) | No clear effect |
| add loss-guided auxiliary agent for valid bit sequences | GFlowNet TB · Valid bit sequences (length 40) | diversity: 89 -> 7746 | Improved |
| add loss-guided auxiliary agent for Bayesian structure learning | GFlowNet TB · Bayesian structure learning (9+ nodes) | E-SHD: 13 nodes: 64.20 -> 63.00; 11 nodes: 42.60 -> 41.33 | Mixed |
| vary lambda in the auxiliary reward (0.0 to 1.5) (ablation) | LGGFN · Hypergrid 32x32 and 64x64 | L1 loss | No clear effect |
| add loss-guided auxiliary agent to mRNA design | GFlowNet TB · mRNA design ACOT13 | CAI: CAI up to 0.62 vs 0.54 natural | Improved |
Closest papers by meaning
- Beyond Squared Error: Exploring Loss Design for Enhanced Training of Generative Flow Networks (ICLR 2025 · similarity 0.91)
- Pessimistic Backward Policy for GFlowNets (NeurIPS 2024 · similarity 0.90)
- Pre-Training and Fine-Tuning Generative Flow Networks (ICLR 2024 · similarity 0.90)
- Order-Preserving GFlowNets (ICLR 2024 · similarity 0.89)
- gfnx: Fast and Scalable Library for Generative Flow Networks in JAX (NeurIPS 2026 · similarity 0.89)
- Flow Factorization for Efficient Generative Flow Networks (AAAI 2025 · similarity 0.89)
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.
