AI Paper Atlas

Research map / Reinforcement learning / Offline RL

Beyond the Proxy: Trajectory-Distilled Guidance for Offline GFlowNet Training

Ruishuo Chen, Xun Wang, Rui Hu, Zhuoran Li, Longbo Huang

ICML 2026 · Reinforcement learning › Offline RL · approach: Diffusion models for offline decision making · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
use TD-GFN instead of QM-COFlowNetGFlowNet offline training · Molecule DesignReward-10: 7.611 -> 7.733Improved
use TD-GFN instead of QM-COFlowNetGFlowNet offline training · Molecule DesignConvergence (trajectories): 4.423 x 10^4 -> 2.749 x 10^4Improved
use TD-GFN instead of Proxy-GFNGFlowNet offline training · Molecule DesignReward-1000: 6.636 -> 6.810Improved
use TD-GFN instead of Oracle-GFN trainingGFlowNet · Molecule DesignConvergence (trajectories): 44.141 x 10^4 -> 2.749 x 10^4Improved
use TD-GFN instead of Dataset-GFNoffline GFlowNet · Molecule DesignReward-100: 7.198 -> 7.450Improved
use TD-GFN instead of Behavior-GFN policyMolecule Design data · Molecule DesignReward-1000: 6.504 -> 6.810Improved
use BraVE offline RL instead of TD-GFNoffline method · Molecule DesignConvergence (trajectories): 2.749 x 10^4 -> 0.645 x 10^4Mixed
use TD-GFN instead of strongest baselines for mode discoveryoffline GFlowNet · Molecule Design (high-reward modes)number of modes: 1.5 to 2 times more modesImproved
use TD-GFN instead of baselinesoffline GFlowNet · 8^4 Hypergridstate visits to find all 16 modes: fewer than 5,000 state visits; six-fold improvementImproved
use TD-GFN instead of COFlowNetoffline GFlowNet · 8^4 Hypergridspeed to lowest Empirical L1 Error: over twice as fastImproved
train GAIL on the rebalanced dataset (ablation)GAIL · 8^4 HypergridperformanceImproved
use rebalanced GAIL instead of Dataset-GFNBiosequence AMP design · AMP designreward and diversity of top 100Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy