AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

All Circuits Lead to Rome: Rethinking Functional Anisotropy in Circuit and Sheaf Discovery for LLMs

Xi Chen, Mingyu Jin, Jingcheng Niu, Yutong Yin, Jinman Zhao, Bangwei Guo, Dimitris N. Metaxas, Zhaoran Wang, Yutao Yue, Gerald Penn

ICML 2026 · Large language models › Language, safety and interpretability · approach: Reasoning mechanisms and failure modes · 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
add overlap-aware sheaf repulsion penalty across discovery runs (ablation)DiscoGP · IOI, 20 sheavesmutual IoU: 0.30% -> 0.15%Improved
add overlap-aware sheaf repulsion penalty across discovery runs (ablation)DiscoGP · IOI, 20 sheavesaverage accuracy: 99.95 -> 99.59No clear effect
add overlap-aware sheaf repulsion penalty across discovery runs (ablation)DiscoGP · BLiMP, 20 sheavesmutual IoU: 1.03% -> 0.70%Improved
add overlap-aware sheaf repulsion penalty across discovery runs (ablation)DiscoGP · ANA, 20 sheavesmutual IoU: 0.57% -> 0.20%Improved
add overlap-aware sheaf repulsion penalty across discovery runs (ablation)DiscoGP · DNA, 20 sheavesmutual IoU: 0.54% -> 0.49%No clear effect
increase number of discovered sheaves from two to 20 (ablation)DiscoGP with OASR · IOIoverlap (IoU): pairwise 4.1% -> mutual 0.15%Hurt
vary attention head traversal order in ACDC (ablation)ACDC · IOIcircuit stability: interchange |E∩|=174 vs |E∪|=608Hurt
use zero ablation instead of interchange ablation (ablation)ACDC · IOIedge density: 1.09% -> 6.76%Hurt
vary only names in IOI prompts, increasing top-k edges (ablation)EAP · IOIpairwise IoU between circuitsHurt
replace KL-divergence loss with task-specific cross entropy loss (ablation)Edge Pruning · IOImin pairwise IoU: 0.340 -> 0.064Hurt
replace KL-divergence loss with task-specific cross entropy loss (ablation)Edge Pruning · Docstringaccuracy: 88.25 -> 99.75Improved
exclude the three-edge core from sheaf discovery on whole IOI (ablation)DiscoGP on IOI · IOIaccuracy: to 52.3%Hurt

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