ICCU: In-Context Continual Unlearning via Pattern-Induced Refusal Rules
The paper introduces ICCU, a framework for in-context continual unlearning in machine learning. It addresses challenges in removing specific data from language models without incurring high costs or interference from multiple unlearning requests. The proposed method effectively suppresses target knowledge while maintaining model utility and robustness across various queries.
- ▪ICCU stands for In-Context Continual Unlearning and uses pattern-induced refusal rules.
- ▪The framework allows for unlearning without modifying model parameters, thus reducing costs and interference.
- ▪Extensive experiments demonstrate ICCU's effectiveness in suppressing knowledge while preserving utility.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.27138 (cs) [Submitted on 26 May 2026] Title:ICCU: In-Context Continual Unlearning via Pattern-Induced Refusal Rules Authors:Ruihao Pan, Suhang Wang View a PDF of the paper titled ICCU: In-Context Continual Unlearning via Pattern-Induced Refusal Rules, by Ruihao Pan and 1 other authors View PDF HTML (experimental) Abstract:Machine unlearning aims to remove the influence of specific data from trained language models. In real-world deployments, unlearning requests often arrive sequentially, which challenges existing fine-tuning-based methods: fine-tuning each request is costly, accumulates utility loss, and may cause cross-request interference.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
