AI Paper Atlas

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

ICML 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Sparse concepts and hypothesis testing · 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
weight forward policies by reaching probability instead of uniform ensemble mixing (ablation)GFlowNet mixing · 2D grid scalarization (2 objectives)L1 error: 0.117 -> 0.003Improved
weight forward policies by reaching probability instead of uniform ensemble mixing (ablation)GFlowNet mixing · 2D grid logical: Circle2 contrast Circle1L1 error: 0.390 -> 0.070Improved
increase number of objectives from 2 to 5Ours · 2D grid scalarizationL1 error: 0.003 -> 0.003No clear effect
increase number of objectives from 2 to 5MOGFN · 2D grid scalarizationL1 error: 0.021 -> 0.048Hurt
increase number of objectives from 2 to 5HN-GFN · 2D grid scalarizationL1 error: 0.017 -> 0.035Hurt
compose pre-trained GFlowNets instead of classifier guidancelogical operators · 2D grid logical: Shubert harmonic mean DiagonalL1 error: 0.142 -> 0.180Hurt
compose pre-trained GFlowNets instead of classifier guidancelogical operators · 2D grid logical: Circle1 harmonic mean Circle2L1 error: 0.397 -> 0.229Improved
compose pre-trained GFlowNets instead of classifier guidancethree-objective harmonic mean · Fragment logical SEH x SA x QEDTarget-bin samples (%): 40 -> 66Improved
compose pre-trained GFlowNets instead of classifier guidancelogical operators · Fragment logical generation timeGeneration time (ms): 1900.71 -> 29.62Improved
compose pre-trained GFlowNets instead of preference-conditioned MOGFNscalarization · QM9 ALLTop-10 reward: 0.688 -> 0.727Improved
compose pre-trained GFlowNets instead of HN-GFNscalarization · Fragment ALLHypervolume: 0.651 -> 0.674Improved
compose pre-trained GFlowNets instead of preference-conditioned baselinesscalarization · Fragment SEH-SATop-10 reward: 0.839 -> 0.836No clear effect

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