Copertina

Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Wei, Boyi, Huang, Kaixuan, Huang, Yangsibo, Xie, Tinghao, Qi, Xiangyu, Xia, Mengzhou, Mittal, Prateek, Wang, Mengdi, Henderson, Peter
Natura:	Preprint
Pubblicazione:	2024
Soggetti:	Machine Learning Artificial Intelligence Computation and Language
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

On Evaluating the Durability of Safeguards for Open-Weight LLMs
di: Qi, Xiangyu, et al.
Pubblicazione: (2024)

SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusal
di: Xie, Tinghao, et al.
Pubblicazione: (2024)

An Adversarial Perspective on Machine Unlearning for AI Safety
di: Łucki, Jakub, et al.
Pubblicazione: (2024)

AI Risk Management Should Incorporate Both Safety and Security
di: Qi, Xiangyu, et al.
Pubblicazione: (2024)

Safety Alignment Should Be Made More Than Just a Few Tokens Deep
di: Qi, Xiangyu, et al.
Pubblicazione: (2024)

What is in Your Safe Data? Identifying Benign Data that Breaks Safety
di: He, Luxi, et al.
Pubblicazione: (2024)

Fantastic Copyrighted Beasts and How (Not) to Generate Them
di: He, Luxi, et al.
Pubblicazione: (2024)

Evaluating Copyright Takedown Methods for Language Models
di: Wei, Boyi, et al.
Pubblicazione: (2024)

SpecDec++: Boosting Speculative Decoding via Adaptive Candidate Lengths
di: Huang, Kaixuan, et al.
Pubblicazione: (2024)

The Model Hears You: Audio Language Model Deployments Should Consider the Principle of Least Privilege
di: He, Luxi, et al.
Pubblicazione: (2025)

Averaging quadratically twisted $L$-values and their derivatives
di: Huang, Tinghao
Pubblicazione: (2025)

Sheared LLaMA: Accelerating Language Model Pre-training via Structured Pruning
di: Xia, Mengzhou, et al.
Pubblicazione: (2023)

Detecting Pretraining Data from Large Language Models
di: Shi, Weijia, et al.
Pubblicazione: (2023)

On Ramanujan Primes for Hecke-Maass Cusp Forms
di: Huang, Tinghao, et al.
Pubblicazione: (2026)

Interpretable Safety Alignment via SAE-Constructed Low-Rank Subspace Adaptation
di: Wang, Dianyun, et al.
Pubblicazione: (2025)

A Theoretical Perspective for Speculative Decoding Algorithm
di: Yin, Ming, et al.
Pubblicazione: (2024)

Adaptive and Stratified Subsampling for High-Dimensional Robust Estimation
di: Mittal, Prateek, et al.
Pubblicazione: (2024)

Lottery Ticket Adaptation: Mitigating Destructive Interference in LLMs
di: Panda, Ashwinee, et al.
Pubblicazione: (2024)

SaLoRA: Safety-Alignment Preserved Low-Rank Adaptation
di: Li, Mingjie, et al.
Pubblicazione: (2025)

A Shear‐Deformable Extended Quasi‐Bond Method With Dual‐Mechanism Fracture Criterion for Brittle and Quasi‐Brittle Materials
di: Wei‐Tong Li, et al.
Pubblicazione: (2025)

Jointly Training and Pruning CNNs via Learnable Agent Guidance and Alignment
di: Ganjdanesh, Alireza, et al.
Pubblicazione: (2024)

LSSF: Safety Alignment for Large Language Models through Low-Rank Safety Subspace Fusion
di: Zhou, Guanghao, et al.
Pubblicazione: (2026)

PRILoRA: Pruned and Rank-Increasing Low-Rank Adaptation
di: Benedek, Nadav, et al.
Pubblicazione: (2024)

Think in Safety: Unveiling and Mitigating Safety Alignment Collapse in Multimodal Large Reasoning Model
di: Lou, Xinyue, et al.
Pubblicazione: (2025)

Advancing LLM Safe Alignment with Safety Representation Ranking
di: Du, Tianqi, et al.
Pubblicazione: (2025)

Diffusion Model for Manifold Data: Score Decomposition, Curvature, and Statistical Complexity
di: Zhang, Zixuan, et al.
Pubblicazione: (2026)

Pruning Long Chain-of-Thought of Large Reasoning Models via Small-Scale Preference Optimization
di: Hong, Bin, et al.
Pubblicazione: (2025)

Immune: Improving Safety Against Jailbreaks in Multi-modal LLMs via Inference-Time Alignment
di: Ghosal, Soumya Suvra, et al.
Pubblicazione: (2024)

When Context Flips, Safety Breaks: Diagnosing Brittle Safety in Aligned Language Models
di: Choi, Dasol, et al.
Pubblicazione: (2026)

Safe Pruning LoRA: Robust Distance-Guided Pruning for Safety Alignment in Adaptation of LLMs
di: Ao, Shuang, et al.
Pubblicazione: (2025)

PatchDEMUX: A Certifiably Robust Framework for Multi-label Classifiers Against Adversarial Patches
di: Jacob, Dennis, et al.
Pubblicazione: (2025)

Beyond Images: Adaptive Fusion of Visual and Textual Data for Food Classification
di: Mittal, Prateek, et al.
Pubblicazione: (2023)

Efficient Data Shapley for Weighted Nearest Neighbor Algorithms
di: Wang, Jiachen T., et al.
Pubblicazione: (2024)

Enhancing Stability and Safety of Commercial Solid‐State Lithium Batteries Through Ternary Eutectic Solvents for Solid‐State Electrolyte Interface Modification
di: Kaixuan Zhou, et al.
Pubblicazione: (2024)

Data Pruning by Information Maximization
di: Tan, Haoru, et al.
Pubblicazione: (2025)

Latent Diffusion Models for Controllable RNA Sequence Generation
di: Huang, Kaixuan, et al.
Pubblicazione: (2024)

Emergent Symbolic Mechanisms Support Abstract Reasoning in Large Language Models
di: Yang, Yukang, et al.
Pubblicazione: (2025)

Embarrassment and the Social Dimensions of Moral Agency
di: Shawn Tinghao Wang
Pubblicazione: (2026)

The finite cohesiveness principle
di: Sun, Mengzhou
Pubblicazione: (2025)

The Kaufmann--Clote question on end extensions of models of arithmetic and the weak regularity principle
di: Sun, Mengzhou
Pubblicazione: (2024)