Kevin Kuo, Amrith Setlur, Kartik Srinivas, Aditi Raghunathan, Virginia Smith
Approximate unlearning has gained popularity as an approach to efficiently update an LLM so that it behaves (roughly) as if it was not trained on a subset of data to begin with. However, existing methods are brittle in practice and can easily be attacked to reveal supposedly unlearned information. To alleviate issues with approximate unlearning, we instead propose SIFT-Masks (SIgn-Fixed Tuning-Masks), an exact unlearning method based on model merging. SIFT-Masks addresses two key limitations of standard model merging: (1) merging a large number of tasks can severely harm utility; and (2) methods that boost utility by sharing extra information across tasks make exact unlearning prohibitively expensive. SIFT-Masks solves these issues by (1) applying local masks to recover task-specific performance; and (2) constraining finetuning to align with a global sign vector as a lightweight approach to determine masks independently before merging. Across four settings where we merge up to 500 models, SIFT-Masks improves accuracy by 5-80% over naive merging and uses up to 250x less compute for exact unlearning compared to other merging baselines.
Xiangchen Song Carnegie Mellon University, Aashiq Muhamed, Yujia Zheng Carnegie Mellon University, Lingjing Kong Computer Science Department, School of Computer Science, Zeyu Tang Stanford University, Mona T. Diab Carnegie Mellon University, Virginia Smith Carnegie Mellon University, Kun Zhang Carnegie Mellon University and Mohamed bin Zayed University of Artificial Intelligence
Sparse Autoencoders (SAEs) are a prominent tool in mechanistic interpretability (MI) for decomposing neural network activations into interpretable features. However, the aspiration to identify a canonical set of features is challenged by the observed inconsistency of learned SAE features across different training runs, undermining reproducibility and complicating model comparison. We study run-to-run feature consistency in SAEs and argue that it should be reported as a standard evaluation axis alongside reconstruction and sparsity. We propose the Pairwise Dictionary Mean Correlation Coefficient (PW-MCC) as an assignment-based metric to quantify consistency and demonstrate that high levels are achievable (PW-MCC ≈ 0.80 for TopK SAEs on LLM activations) with appropriate architectural choices.Our contributions include: (i) theoretical grounding for strong consistency in the idealized setting of TopK SAEs; (ii) synthetic validation using a model organism, which verifies PW-MCC as a reliable proxy for ground-truth recovery; and (iii) empirical analysis on LLM activations, where PW-MCC correlates with the similarity of automatically generated natural-language feature explanations.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表