Image de couverture de livre

SAEs $\textit{Can}$ Improve Unlearning: Dynamic Sparse Autoencoder Guardrails for Precision Unlearning in LLMs

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Muhamed, Aashiq, Bonato, Jacopo, Diab, Mona, Smith, Virginia
Format:	Preprint
Publié:	2025
Sujets:	Machine Learning Artificial Intelligence Computation and Language Cryptography and Security
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Decoding Dark Matter: Specialized Sparse Autoencoders for Interpreting Rare Concepts in Foundation Models
par: Muhamed, Aashiq, et autres
Publié: (2024)

CoRAG: Collaborative Retrieval-Augmented Generation
par: Muhamed, Aashiq, et autres
Publié: (2025)

DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment
par: Wedgwood, James, et autres
Publié: (2026)

Position: Mechanistic Interpretability Should Prioritize Feature Consistency in SAEs
par: Song, Xiangchen, et autres
Publié: (2025)

Unlearning Isn't Deletion: Investigating Reversibility of Machine Unlearning in LLMs
par: Xu, Xiaoyu, et autres
Publié: (2025)

Towards Robust Knowledge Unlearning: An Adversarial Framework for Assessing and Improving Unlearning Robustness in Large Language Models
par: Yuan, Hongbang, et autres
Publié: (2024)

Textual Unlearning Gives a False Sense of Unlearning
par: Du, Jiacheng, et autres
Publié: (2024)

Unlearned but Not Forgotten: Data Extraction after Exact Unlearning in LLM
par: Wu, Xiaoyu, et autres
Publié: (2025)

UnUnlearning: Unlearning is not sufficient for content regulation in advanced generative AI
par: Shumailov, Ilia, et autres
Publié: (2024)

Personal Information Parroting in Language Models
par: Subramani, Nishant, et autres
Publié: (2026)

An Adversarial Perspective on Machine Unlearning for AI Safety
par: Łucki, Jakub, et autres
Publié: (2024)

Unlearning Backdoor Attacks for LLMs with Weak-to-Strong Knowledge Distillation
par: Zhao, Shuai, et autres
Publié: (2024)

Unlink to Unlearn: Simplifying Edge Unlearning in GNNs
par: Tan, Jiajun, et autres
Publié: (2024)

Machine Unlearning of Pre-trained Large Language Models
par: Yao, Jin, et autres
Publié: (2024)

LLM Ghostbusters: Surgical Hallucination Suppression via Adaptive Unlearning
par: Spracklen, Joseph, et autres
Publié: (2026)

FIT to Forget: Robust Continual Unlearning for Large Language Models
par: Xu, Xiaoyu, et autres
Publié: (2026)

OBLIVIATE: Robust and Practical Machine Unlearning for Large Language Models
par: Xu, Xiaoyu, et autres
Publié: (2025)

Fast Exact Unlearning for In-Context Learning Data for LLMs
par: Muresanu, Andrei I., et autres
Publié: (2024)

Machine Unlearning with Minimal Gradient Dependence for High Unlearning Ratios
par: Huang, Tao, et autres
Publié: (2024)

From Domains to Instances: Dual-Granularity Data Synthesis for LLM Unlearning
par: Xu, Xiaoyu, et autres
Publié: (2026)

RefusalBench: Generative Evaluation of Selective Refusal in Grounded Language Models
par: Muhamed, Aashiq, et autres
Publié: (2025)

GraphToxin: Reconstructing Full Unlearned Graphs from Graph Unlearning
par: Song, Ying, et autres
Publié: (2025)

Shadow Unlearning: A Neuro-Semantic Approach to Fidelity-Preserving Faceless Forgetting in LLMs
par: P, Dinesh Srivasthav, et autres
Publié: (2026)

Towards Understanding the Robustness of Sparse Autoencoders
par: Saiyed, Ahson, et autres
Publié: (2026)

Towards Lifecycle Unlearning Commitment Management: Measuring Sample-level Unlearning Completeness
par: Wang, Cheng-Long, et autres
Publié: (2025)

Data-Free Privacy-Preserving for LLMs via Model Inversion and Selective Unlearning
par: Zhou, Xinjie, et autres
Publié: (2026)

Recalling The Forgotten Class Memberships: Unlearned Models Can Be Noisy Labelers to Leak Privacy
par: Sui, Zhihao, et autres
Publié: (2025)

DUP: Detection-guided Unlearning for Backdoor Purification in Language Models
par: Hu, Man, et autres
Publié: (2025)

Prompt Attacks Reveal Superficial Knowledge Removal in Unlearning Methods
par: Jang, Yeonwoo, et autres
Publié: (2025)

CCRS: A Zero-Shot LLM-as-a-Judge Framework for Comprehensive RAG Evaluation
par: Muhamed, Aashiq
Publié: (2025)

Backdoor Token Unlearning: Exposing and Defending Backdoors in Pretrained Language Models
par: Jiang, Peihai, et autres
Publié: (2025)

Harry Potter is Still Here! Probing Knowledge Leakage in Targeted Unlearned Large Language Models via Automated Adversarial Prompting
par: To, Bang Trinh Tran, et autres
Publié: (2025)

Unlearning Inversion Attacks for Graph Neural Networks
par: Zhang, Jiahao, et autres
Publié: (2025)

SoK: Unlearnability and Unlearning for Model Dememorization
par: Zhang, Mengying, et autres
Publié: (2026)

Forgetting-MarI: LLM Unlearning via Marginal Information Regularization
par: Xu, Shizhou, et autres
Publié: (2025)

ThinkGuard: Deliberative Slow Thinking Leads to Cautious Guardrails
par: Wen, Xiaofei, et autres
Publié: (2025)

Pando: Do Interpretability Methods Work When Models Won't Explain Themselves?
par: Zhong, Ziqian, et autres
Publié: (2026)

In-Context Unlearning: Language Models as Few Shot Unlearners
par: Pawelczyk, Martin, et autres
Publié: (2023)

WARP: Weight Teleportation for Attack-Resilient Unlearning Protocols
par: Maheri, Mohammad M, et autres
Publié: (2025)

Scalable Federated Unlearning via Isolated and Coded Sharding
par: Lin, Yijing, et autres
Publié: (2024)