ICCU: In-Context Continual Unlearning via Pattern-Induced Refusal Rules

📰 ArXiv cs.AI

Learn how to implement ICCU, a framework for in-context continual unlearning in language models, to efficiently remove data influence without fine-tuning

advanced Published 27 May 2026

Action Steps

Implement ICCU framework using pattern-induced refusal rules to identify data to be unlearned
Configure the ICCU model to receive sequential unlearning requests
Apply in-context learning to update the model without fine-tuning
Test the ICCU model for utility loss and cross-request interference
Compare ICCU performance with fine-tuning-based methods

Who Needs to Know This

Machine learning engineers and researchers working on language models can benefit from ICCU to improve model performance and adapt to changing data requirements

Key Insight

💡 ICCU enables sequential unlearning without fine-tuning, reducing utility loss and cross-request interference

Key Takeaways

Learn how to implement ICCU, a framework for in-context continual unlearning in language models, to efficiently remove data influence without fine-tuning

Full Article

Title: ICCU: In-Context Continual Unlearning via Pattern-Induced Refusal Rules

Abstract:
arXiv:2605.27138v1 Announce Type: new 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. To address these issues, we propose ICCU (In-Context Continual Unlearning), an in-context continual unlearning framework that

Read full paper → ← Back to Reads