SCAN: Sparse Circuit Anchor Interpretable Neuron for Lifelong Knowledge Editing

TL;DR

SCAN introduces a sparse, mechanism-aware editing framework for LLMs that significantly reduces catastrophic forgetting during sequential knowledge updates, outperforming existing dense editing methods.

Contribution

The paper presents SCAN, a novel sparse circuit-based editing approach that enhances knowledge editing precision and model stability in large language models.

Findings

SCAN maintains model performance after 3,000 sequential edits.

It outperforms existing methods on benchmarks like MMLU and GSM8K.

Model collapse is mitigated with the proposed sparse editing framework.

Abstract

Large Language Models (LLMs) often suffer from catastrophic forgetting and collapse during sequential knowledge editing. This vulnerability stems from the prevailing dense editing paradigm, which treats models as black boxes and relies on coarse-grained parameter interventions that inevitably disrupt preserved knowledge. To address this, we propose SCAN (a sparse editing framework based on Sparse Circuit Anchored Neuron) which transforms editing into a mechanism-aware manipulation by constructing a knowledge circuit via Sparse Transcoders. Experiments on Gemma2, Qwen3, and Llama3.1 across CounterFact, ZsRE and WikiFactDiff demonstrate that SCAN achieves a superior performance, maintaining model integrity on benchmarks like MMLU and GSM8K even after 3,000 sequential edits, whereas other existing methods deteriorate progressively as editing accumulates, eventually resulting in model collapse.