Academic

U-CAN: Utility-Aware Contrastive Attenuation for Efficient Unlearning in Generative Recommendation

arXiv:2602.23400v1 Announce Type: new Abstract: Generative Recommendation (GenRec) typically leverages Large Language Models (LLMs) to redefine personalization as an instruction-driven sequence generation task. However, fine-tuning on user logs inadvertently encodes sensitive attributes into model parameters, raising critical privacy concerns. Existing Machine Unlearning (MU) techniques struggle to navigate this tension due to the Polysemy Dilemma, where neurons superimpose sensitive data with general reasoning patterns, leading to catastrophic utility loss under traditional gradient or pruning methods. To address this, we propose Utility-aware Contrastive AttenuatioN (U-CAN), a precision unlearning framework that operates on low-rank adapters. U-CAN quantifies risk by contrasting activations and focuses on neurons with asymmetric responses that are highly sensitive to the forgetting set but suppressed on the retention set. To safeguard performance, we introduce a utility-aware calibr

Z
Zezheng Wu, Rui Wang, Xinghe Cheng, Yang Shao, Qing Yang, Jiapu Wang, Jingwei Zhang · · 1 min read · 9 views
#cs.LG

arXiv:2602.23400v1 Announce Type: new Abstract: Generative Recommendation (GenRec) typically leverages Large Language Models (LLMs) to redefine personalization as an instruction-driven sequence generation task. However, fine-tuning on user logs inadvertently encodes sensitive attributes into model parameters, raising critical privacy concerns. Existing Machine Unlearning (MU) techniques struggle to navigate this tension due to the Polysemy Dilemma, where neurons superimpose sensitive data with general reasoning patterns, leading to catastrophic utility loss under traditional gradient or pruning methods. To address this, we propose Utility-aware Contrastive AttenuatioN (U-CAN), a precision unlearning framework that operates on low-rank adapters. U-CAN quantifies risk by contrasting activations and focuses on neurons with asymmetric responses that are highly sensitive to the forgetting set but suppressed on the retention set. To safeguard performance, we introduce a utility-aware calibration mechanism that combines weight magnitudes with retention-set activation norms, assigning higher utility scores to dimensions that contribute strongly to retention performance. Unlike binary pruning, which often fragments network structure, U-CAN develop adaptive soft attenuation with a differentiable decay function to selectively down-scale high-risk parameters on LoRA adapters, suppressing sensitive retrieval pathways and preserving the topological connectivity of reasoning circuits. Experiments on two public datasets across seven metrics demonstrate that U-CAN achieves strong privacy forgetting, utility retention, and computational efficiency.

Executive Summary

This article proposes Utility-aware Contrastive AttenuatioN (U-CAN), a novel unlearning framework designed to address privacy concerns in Generative Recommendation (GenRec) systems. U-CAN leverages low-rank adapters to selectively attenuate high-risk parameters while preserving utility. By contrasting activations and utilizing a utility-aware calibration mechanism, U-CAN achieves strong privacy forgetting, utility retention, and computational efficiency. The authors demonstrate the effectiveness of U-CAN on two public datasets across seven metrics. However, further research is needed to fully explore the generalizability and applicability of U-CAN to real-world GenRec systems.

Key Points

  • ▸ U-CAN framework operates on low-rank adapters to selectively attenuate high-risk parameters.
  • ▸ Utility-aware calibration mechanism preserves utility while prioritizing privacy.
  • ▸ Experiments demonstrate strong privacy forgetting, utility retention, and computational efficiency.

Merits

Strength in addressing the Polysemy Dilemma

U-CAN effectively navigates the tension between privacy and utility by selectively attenuating high-risk parameters.

Demerits

Limited generalizability

Further research is needed to fully explore the applicability of U-CAN to real-world GenRec systems and diverse datasets.

Expert Commentary

The proposed U-CAN framework represents a significant advancement in addressing the Polysemy Dilemma in GenRec systems. By leveraging low-rank adapters and a utility-aware calibration mechanism, U-CAN effectively balances privacy and utility, providing a promising solution for real-world applications. However, the authors should be encouraged to further explore the generalizability and applicability of U-CAN to diverse datasets and real-world scenarios. This would enable a more comprehensive understanding of the framework's capabilities and limitations.

Recommendations

  • ✓ Future research should focus on exploring the generalizability and applicability of U-CAN to diverse datasets and real-world scenarios.
  • ✓ The authors should investigate the potential of U-CAN in other domains beyond GenRec, such as natural language processing and computer vision.

Sources

Related Articles