LGMay 14, 2025

Layered Unlearning for Adversarial Relearning

arXiv:2505.09500v11 citationsh-index: 8Has Code
Originality Incremental advance
AI Analysis

This work addresses the brittleness of post-training updates in language models, which is a critical issue for ensuring reliable model behavior, though it appears incremental in the context of machine unlearning.

The paper tackles the problem of post-training modifications in language models being brittle and easily bypassed, and introduces Layered Unlearning (LU) to improve robustness against adversarial relearning, showing enhanced performance across various unlearning methods.

Our goal is to understand how post-training methods, such as fine-tuning, alignment, and unlearning, modify language model behavior and representations. We are particularly interested in the brittle nature of these modifications that makes them easy to bypass through prompt engineering or relearning. Recent results suggest that post-training induces shallow context-dependent ``circuits'' that suppress specific response patterns. This could be one explanation for the brittleness of post-training. To test this hypothesis, we design an unlearning algorithm, Layered Unlearning (LU), that creates distinct inhibitory mechanisms for a growing subset of the data. By unlearning the first $i$ folds while retaining the remaining $k - i$ at the $i$th of $k$ stages, LU limits the ability of relearning on a subset of data to recover the full dataset. We evaluate LU through a combination of synthetic and large language model (LLM) experiments. We find that LU improves robustness to adversarial relearning for several different unlearning methods. Our results contribute to the state-of-the-art of machine unlearning and provide insight into the effect of post-training updates.

Code Implementations1 repo
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes