AI Paper Atlas

Research map / Probabilistic and scientific modeling / Causal discovery

Mechanistic Interpretability as Statistical Estimation: A Variance Analysis

Maxime Méloux, François Portet, Maxime Peyrard

ICML 2026 · Probabilistic and scientific modeling › Causal discovery · approach: Latent-variable causal estimation · 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 bootstrap resampling instead of meta-dataset resamplingEAP-IG circuit discovery · averaged over models and taskspairwise Jaccard mean: 0.790 -> 0.561Hurt
use bootstrap resampling instead of prompt paraphrasingEAP-IG circuit discovery · averaged over models and taskscircuit error: 0.150 -> 0.440Hurt
replace exact CMA scores with EAP estimatesedge importance scoring · gpt2-small Greater-Thancoefficient of variation of edge scores: CV≈0.5 -> CV>1Hurt
scale from gpt2-small to Llama-3.2 (1B and Instruct)circuit discovery · IOI, SVA, Greater-Thanstability of circuitsHurt
instruction-tune Llama-3.2-1Bcircuit discovery · IOI, SVA, Greater-Thancircuit stabilityNo clear effect
change aggregation sum to median and mean to patching for EAP-IG-inputs (ablation)EAP-IG-inputs · Greater-Than (Llama-3.2-1B-Instruct)Jaccard to median circuit: 1.000 -> 0.086Hurt
swap EAP-IG-inputs for Clean-corrupted (ablation)EAP-IG-inputs, sum, patching · IOI (Llama-3.2-1B-Instruct)Jaccard to median circuit: 0.182 -> 0.071Hurt
use mean-positional instead of patching counterfactual (ablation)EAP-IG-inputs, sum · Greater-Than (Llama-3.2-1B-Instruct)circuit error: 0.20 -> 0.41Hurt
use EAP-IG-activations instead of EAP-IG-inputs (ablation)EAP-IG, sum, patching · Greater-Than (Llama-3.2-1B-Instruct)Jaccard to median circuit: 0.571 -> 0.098Hurt
increase Gaussian noise amplitude of the counterfactual (ablation)EAP-IG circuit discovery · gpt2-smallJaccard index CVMixed
use Clean-corrupted instead of EAP (ablation)circuit size · IOI (Llama-3.2-1B)circuit size (edges): 7 -> 448Hurt

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