AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

LoRe: Adaptive Interaction-Evaluation Routing with Per-step Interaction Budgets for Iterative Graph Solvers

Jintao Li, Yong-Yi Wang, Zheng-an Wang, Heng Fan

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Scalable approximation and equivariant compute · 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 dynamic state-dependent re-routing instead of a static interaction support (LoRe-Static) (ablation)LoRe on MIS conflict-descent relaxation · 40 ER/BA graphs, rho=0.08conflict convergence and legal solution sizeImproved
combine a static skeleton with dynamic hotspots rather than purely dynamic greedy edge selection (ablation)LoRe routing · 40 ER/BA graphs, rho=0.08stability of convergenceImproved
sweep proxy-score weight lambda_stab from 0 to 5.0 (ablation)LoRe · DIFUSCO MIS n=5000quality retention: range 0.9 pp / 2.8 pp / 1.6 pp across densitiesNo clear effect
vary interaction budget rho in [0.05, 0.50] (ablation)LoRe · MISquality retention: less than 3 pp variationNo clear effect
vary refresh interval R in [1, 50] (ablation)LoRe · MISquality retention: less than 1.5 pp variationNo clear effect
apply LoRe per-step interaction budgeting to DIFUSCO MIS at larger graph sizeDIFUSCO · MIS n=15kpeak memory (GB): 86.7 -> 7.32Improved
apply LoRe to DIFUSCO MIS beyond baseline OOM limitDIFUSCO · MIS n=20k-50kfeasibility: baseline OOM; LoRe feasible up to n=50kImproved
apply LoRe to DIFUSCO TSP at small nDIFUSCO · TSP n=100end-to-end time (s): 0.34 -> 0.61Hurt
apply LoRe to DIFUSCO TSP at n=1000DIFUSCO · TSP n=1kend-to-end time (s): 13.6 -> 0.94Improved
apply LoRe zero-shot to scale-free BA graphs with ER-pretrained modelDIFUSCO · MIS topology OODquality retention: 0.991 -> 0.756 (ER vs BA)Hurt
apply LoRe to small MIS instancesDIFUSCO · MIS n=1k-3kquality retention: 0.815, 0.804, 0.835Hurt
apply LoRe to COExpander MISCOExpander · MIS V=2.4kquality retention: 0.872 retentionHurt

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