paper-with-me

홈 › Papers

Continual Safety Alignment via Gradient-Based Sample Selection

2026-04-19 · Thong Bach, Dung Nguyen, Thao Minh Le, Truyen Tran arxiv

Large language models require continuous adaptation to new tasks while preserving safety alignment. However, fine-tuning on even benign data often compromises safety behaviors, including refusal of harmful requests, truthfulness, and commonsense reasoning. We investigate which training samples cause alignment drift through a data-centric lens. Our empirical analysis shows samples contribute unequally: high-gradient samples cause greater safety degradation and drive models toward pretrained distributions, while moderate-gradient samples enable task learning with minimal alignment loss. We propose gradient-based sample selection that filters high-gradient samples during fine-tuning. Across multiple model families on continual domain tasks, our method substantially improves alignment preservation while maintaining competitive task performance, without requiring curated safe data or architectural modifications. Our method is robust across selection ratios, task orderings, and diverse attack benchmarks.

📄 PDF Abstract BibTeX arXiv:2604.17215

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Two to Tango: Coupled Task-Reference Selection for Safe LLM Fine-tuning

2026-06-01 · Xinrui Chen, Jianhao Zhang, Ou Wu, Di Gao arxiv

Fine-tuning safety aligned large language models (LLMs) on downstream data improves adaptation but may erode learned safety behavior. Existing methods use fixed safety examples, global constraints, or one-sided task filt…

Continual Learning

Coordinated Replay Sample Selection for Continual Federated Learning

2023-10-23 · Jack Good, Jimit Majmudar, Christophe Dupuy, Jixuan Wang 외

Continual Federated Learning (CFL) combines Federated Learning (FL), the decentralized learning of a central model on a number of client devices that may not communicate their data, and Continual Learning (CL), the learn…

Continual LearningFederated Learning

Safety Alignment as Continual Learning: Mitigating the Alignment Tax via Orthogonal Gradient Projection

2026-02-08 · Guanglong Sun, Siyuan Zhang, Liyuan Wang, Jun Zhu 외 arxiv

Safety post-training can improve the harmfulness and policy compliance of Large Language Models (LLMs), but it may also reduce general utility, a phenomenon often described as the \emph{alignment tax}. We study this trad…

Continual Learning

Balanced Gradient Sample Retrieval for Enhanced Knowledge Retention in Proxy-based Continual Learning

2024-12-19 · Hongye Xu, Jan Wasilewski, Bartosz Krawczyk

Continual learning in deep neural networks often suffers from catastrophic forgetting, where representations for previous tasks are overwritten during subsequent training. We propose a novel sample retrieval strategy fro…

Continual LearningContrastive LearningDiversityRetrieval

Not Just Selection, but Exploration: Online Class-Incremental Continual Learning via Dual View Consistency

2022-01-01 · CVPR 2022 1 · Yanan Gu, Xu Yang, Kun Wei, Cheng Deng

Online class-incremental continual learning aims to learn new classes continually from a never-ending and single-pass data stream, while not forgetting the learned knowledge of old classes. Existing replay-based meth…

Continual Learning