BlackboxNLP-2025 MIB Shared Task: Improving Circuit Faithfulness via Better Edge Selection

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Nikankin, Yaniv, Arad, Dana, Itzhak, Itay, Reusch, Anja, Simhi, Adi, Kesten-Pomeranz, Gal, Belinkov, Yonatan
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912677978177536
author Nikankin, Yaniv
Arad, Dana
Itzhak, Itay
Reusch, Anja
Simhi, Adi
Kesten-Pomeranz, Gal
Belinkov, Yonatan
author_facet Nikankin, Yaniv
Arad, Dana
Itzhak, Itay
Reusch, Anja
Simhi, Adi
Kesten-Pomeranz, Gal
Belinkov, Yonatan
contents One of the main challenges in mechanistic interpretability is circuit discovery, determining which parts of a model perform a given task. We build on the Mechanistic Interpretability Benchmark (MIB) and propose three key improvements to circuit discovery. First, we use bootstrapping to identify edges with consistent attribution scores. Second, we introduce a simple ratio-based selection strategy to prioritize strong positive-scoring edges, balancing performance and faithfulness. Third, we replace the standard greedy selection with an integer linear programming formulation. Our methods yield more faithful circuits and outperform prior approaches across multiple MIB tasks and models. Our code is available at: https://github.com/technion-cs-nlp/MIB-Shared-Task.
format Preprint
id arxiv_https___arxiv_org_abs_2510_25786
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle BlackboxNLP-2025 MIB Shared Task: Improving Circuit Faithfulness via Better Edge Selection
Nikankin, Yaniv
Arad, Dana
Itzhak, Itay
Reusch, Anja
Simhi, Adi
Kesten-Pomeranz, Gal
Belinkov, Yonatan
Computation and Language
Artificial Intelligence
One of the main challenges in mechanistic interpretability is circuit discovery, determining which parts of a model perform a given task. We build on the Mechanistic Interpretability Benchmark (MIB) and propose three key improvements to circuit discovery. First, we use bootstrapping to identify edges with consistent attribution scores. Second, we introduce a simple ratio-based selection strategy to prioritize strong positive-scoring edges, balancing performance and faithfulness. Third, we replace the standard greedy selection with an integer linear programming formulation. Our methods yield more faithful circuits and outperform prior approaches across multiple MIB tasks and models. Our code is available at: https://github.com/technion-cs-nlp/MIB-Shared-Task.
title BlackboxNLP-2025 MIB Shared Task: Improving Circuit Faithfulness via Better Edge Selection
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2510.25786