Safety Alignment as Continual Learning: Mitigating the Alignment Tax via Orthogonal Gradient Projection
School of Life Sciences, IDG/McGovern Institute for Brain Research（生命科学学院，IDG/麦戈文脑科学研究所）；Dept. of Comp. Sci. and Tech., Institute for AI, Tsinghua-Bosch Joint ML Center, THBI Lab, BNRist Center（计算机科学与技术系，人工智能研究所，清华大学-博世联合机器学习中心，THBI实验室，BNRist中心）；Tsinghua University, Beijing, China（清华大学，北京，中国）
Guanglong Sun, Siyuan Zhang, Liyuan Wang, Jun Zhu, Hang Su, Yi Zhong
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 trade-off through the lens of continual learning: sequential alignment stages expose the model to shifted data distributions and objectives, and their gradients may interfere with directions that support previously acquired general capabilities. This view does not claim that all alignment degradation has a single cause; rather, it provides a useful first-order mechanism for mitigating one important source of capability regression. We propose \textbf{O}rthogonal \textbf{G}radient \textbf{P}rojection for \textbf{S}afety \textbf{A}lignment (\textbf{OGPSA}), a lightweight update rule that estimates a low-rank reference subspace from gradients on a small set of general-capability data and removes from each safety gradient the component lying in this subspace. The resulting update is the steepest local safety-descent direction subject to first-order preservation constraints on the reference objectives. OGPSA is compatible with standard post-training pipelines and avoids large-scale replay, although it introduces periodic reference-gradient computation. Across Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and sequential SFT$\rightarrow$DPO settings, OGPSA improves the observed safety--utility trade-off over standard baselines. Under the sequential SFT$\rightarrow$DPO pipeline, the average performance gain increases from 33.98\% to 42.74\% on Qwen2.5-7B-Instruct and from 19.74\% to 32.98\% on Llama3.1-8B-Instruct. We have open sourced our code at https://github.com/SunGL001/OGPSA.
Leveraging Hierarchical Taxonomies in Prompt-based Continual Learning
VinAI Research（VinAI研究院）；New York University（纽约大学）；Monash University（莫纳什大学）；Hanoi University of Science and Technology（河内科技大学）；University of Oregon（俄勒冈大学）；The University of Texas at Austin（德克萨斯大学奥斯汀分校）
PVBF: A Framework for Mitigating Parameter Variation Imbalance in Online Continual Learning
School of Computer Science and Technology, Tongji University（同济大学计算机科学与技术学院）；College of Electronic and Information Engineering, Tongji University（同济大学电子与信息工程学院）；National Key Laboratory for Novel Software Technology, Nanjing University（南京大学计算机软件新技术国家重点实验室）
LongReD: Mitigating Short-Text Degradation of Long-Context Large Language Models via Restoration Distillation
Gaoling School of Artificial Intelligence, Renmin University of China（中国人民大学人工智能学院）；Department of Computer Science, National University of Singapore（新加坡国立大学计算机科学系）；Baichuan Inc.（百川公司）
School of Computer Science and Engineering, Macau University of Science and Technology（澳门科技大学计算机科学与工程学院）；Department of Computer Science, Hong Kong Baptist University（香港浸会大学计算机科学系）
PASs-MoE: Mitigating Misaligned Co-drift among Router and Experts via Pathway Activation Subspaces for Continual Learning
Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所）；University of Chinese Academy of Sciences（中国科学院大学）；National University of Singapore（新加坡国立大学）；Southeast University, Nanjing, China（南京东南大学）；Wuhan AI Research, Wuhan, China（武汉人工智能研究院）
FPPL: An Efficient and Non-IID Robust Federated Continual Learning Framework
School of Computer Science and Technology, East China Normal University（华东师范大学计算机科学与技术学院）；School of Software Engineering, TongJi University（同济大学软件工程学院）
Online Continual Learning with Dynamic Label Hierarchies
College of Computer Science（计算机科学学院）；Technology, Nanjing University of Aeronautics（技术学院，南京航空航天大学）；MIIT Key Laboratory of Pattern Analysis（模式分析 MIIT 实验室）；Computer Vision Center, Spain（西班牙计算机视觉中心）；Computer Sciences Department, Universitat Autonoma de Barcelona, Spain（巴塞罗那自治大学计算机科学系，西班牙）；State Key Laboratory for Novel Software Technology, Nanjing University（新型软件技术国家重点实验室，南京大学）；IDEAS Research Institute, Warsaw, Poland（华沙 IDEAS 研究院，波兰）；Joint Laboratory of Spatial intelligent Perception（空间智能感知联合实验室）
Revisiting Weight Regularization for Low-Rank Continual Learning
State Key Laboratory of Human-Machine Hybrid Augmented Intelligence（人机混合增强智能国家重点实验室）；Institute of Artificial Intelligence and Robotics（人工智能与机器人研究院）；Computer Vision Center（计算机视觉中心）；Universitat Autònoma de Barcelona（巴塞罗那自治大学）；School of Instrument Science and Engineering（仪器科学与工程学院）；Bernoulli Institute（伯努利学院）；University of Groningen（格罗宁根大学）；School of Software Engineering（软件工程学院）
Understanding the Role of Rehearsal Scale in Continual Learning under Varying Model Capacities
Institute of Intelligent Information Processing, Shanxi University（山西大学智能信息处理研究院）；Alliance Manchester Business School, The University of Manchester（曼彻斯特大学阿联酋商业学院）
Mitigating Catastrophic Forgetting in Large Language Models with Self-Synthesized Rehearsal
School of Informatics, Xiamen University（厦门大学信息学院）；Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）；Tencent AI Lab（腾讯AI实验室）；Zhejiang University（浙江大学）
Octopus: History-Free Gradient Orthogonalization for Continual Learning in Multimodal Large Language Models
MoE Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University（人工智能大模型关键实验室，人工智能研究院，上海交通大学）；vivo Mobile Communication Co., Ltd.（vivo移动通信有限公司）
Department of Mathematics, University of Padua（帕多瓦大学数学系）；Faculty of Informatics, Università della Svizzera Italiana（瑞士意大利大学信息学院）；Department of Electronics, Information and Bioengineering, Politecnico di Milano（米兰理工学院电子、信息与生物工程系）
Where is the Truth? The Risk of Getting Confounded in a Continual World
Computer Science Department, TU Darmstadt, Germany.（图宾根大学计算机科学系）；Hessian Center for AI (hessian.AI), Germany.（黑森人工智能中心）；German Research Center for AI (DFKI), Germany（德国人工智能研究中心）；TU Darmstadt, Centre for Cognitive Science, Germany.（图宾根大学认知科学中心）；University of Bremen, Department of Mathematics（不莱梅大学数学系）
