Beyond Retention: Orchestrating Structural Safety and Plasticity in Continual Learning for LLMs

• URL: http://arxiv.org/abs/2601.18255v1
• Date: Mon, 26 Jan 2026 08:28:02 GMT
• Title: Beyond Retention: Orchestrating Structural Safety and Plasticity in Continual Learning for LLMs
• Abstract summary: Experience Replay (ER) is a standard countermeasure against catastrophic forgetting.<n>ER causes negative transfer on fragile, structured domains like code generation.<n>We proposeOrthogonal Subspace Wake-up (OSW) as a mathematically grounded "safety guarantee" for established knowledge structures.
• Score: 2.5202137607634407
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Continual learning in Large Language Models (LLMs) faces the critical challenge of balancing stability (retaining old knowledge) and plasticity (learning new tasks). While Experience Replay (ER) is a standard countermeasure against catastrophic forgetting, its impact across diverse capabilities remains underexplored. In this work, we uncover a critical dichotomy in ER's behavior: while it induces positive backward transfer on robust, unstructured tasks (e.g., boosting performance on previous NLP classification tasks through repeated rehearsal), it causes severe negative transfer on fragile, structured domains like code generation (e.g., a significant relative drop in coding accuracy). This reveals that ER trades structural integrity for broad consolidation. To address this dilemma, we propose \textbf{Orthogonal Subspace Wake-up (OSW)}. OSW identifies essential parameter subspaces of previous tasks via a brief "wake-up" phase and enforces orthogonal updates for new tasks, providing a mathematically grounded "safety guarantee" for established knowledge structures. Empirical results across a diverse four-task sequence demonstrate that OSW uniquely succeeds in preserving fragile coding abilities where Replay fails, while simultaneously maintaining high plasticity for novel tasks. Our findings emphasize the necessity of evaluating structural safety alongside average retention in LLM continual learning.

Related papers

• Robust Dynamic Expansion for Continual Learning under Backdoor Attacks via Purification and Selective Recovery [21.762207770794916]
  In practical scenarios, task streams collected from untrusted sources may contain backdoor-poisoned samples.<n>We propose a robust dynamic-expansion framework that integrates sample purification, selective recovery, and robust expert routing.
  arXiv  Detail & Related papers  (2026-09-06T02:52:44Z)
• RefusalGuard: Geometry-Preserving Fine-Tuning for Safety in LLMs [5.100622189286672]
  Fine-tuning safety-aligned language models for downstream tasks can lead to substantial degradation of refusal behavior.<n>We introduce REFUSALGUARD, a representation-level fine-tuning framework that preserves safety-relevant structure during model adaptation.
  arXiv  Detail & Related papers  (2026-05-03T14:48:18Z)
• Uncovering Linguistic Fragility in Vision-Language-Action Models via Diversity-Aware Red Teaming [64.48633529149579]
  We propose a novel framework to expose the vulnerabilities of Vision-Language-Action (VLA) models against linguistic variations.<n>Our method reduces the average task success rate from 93.33% to 5.85%, demonstrating a scalable approach to stress-testing VLA agents.
  arXiv  Detail & Related papers  (2026-04-07T08:43:36Z)
• ARES: Scalable and Practical Gradient Inversion Attack in Federated Learning through Activation Recovery [73.8181449261685]
  Federated Learning (FL) enables collaborative model training by sharing model updates instead of raw data, aiming to protect user privacy.<n>Recent studies reveal that these shared updates can inadvertently leak sensitive training data through gradient inversion attacks (GIAs)
  arXiv  Detail & Related papers  (2026-03-18T11:40:44Z)
• SPREAD: Subspace Representation Distillation for Lifelong Imitation Learning [11.023696977257883]
  Key challenge in lifelong imitation learning is enabling agents to acquire new skills from expert demonstrations while retaining prior knowledge.<n>Existing distillation methods, which rely on L2-norm feature matching in raw feature space, are sensitive to noise and high-dimensional variability.<n>We introduce SPREAD, a geometry-preserving framework that employs singular value decomposition to align policy representations across tasks within low-rank subspaces.
  arXiv  Detail & Related papers  (2026-03-09T03:38:42Z)
• Spectral Imbalance Causes Forgetting in Low-Rank Continual Adaptation [58.3773038915023]
  Continual learning aims to adapt pre-trained models to sequential tasks without forgetting previously acquired knowledge.<n>Most existing approaches treat continual learning as avoiding interference with past updates, rather than considering what properties make the current task-specific update naturally preserve previously acquired knowledge.<n>We address this problem using a projected first-order method compatible with standard deep-dots used in vision-language models.
  arXiv  Detail & Related papers  (2026-01-31T13:27:02Z)
• Retrofit: Continual Learning with Bounded Forgetting for Security Applications [25.185616916987158]
  We propose RETROFIT, a data retrospective-free continual learning method that achieves bounded forgetting for effective knowledge transfer.<n>To mitigate interference, we apply low-rank and sparse updates that confine parameter changes to independent subspaces.<n>In malware detection under temporal drift, it substantially improves the retention score, from 20.2% to 38.6% over CL baselines, and exceeds the oracle upper bound on new data.
  arXiv  Detail & Related papers  (2025-11-14T16:07:03Z)
• Grounded in Reality: Learning and Deploying Proactive LLM from Offline Logs [72.08224879435762]
  textttLearn-to-Ask is a simulator-free framework for learning and deploying proactive dialogue agents.<n>Our approach culminates in the successful deployment of LLMs into a live, large-scale online AI service.
  arXiv  Detail & Related papers  (2025-10-29T12:08:07Z)
• Spectral Collapse Drives Loss of Plasticity in Deep Continual Learning [14.196969540084929]
  We show that deep neural networks suffer from loss of plasticity in deep continual learning.<n>We introduce the notion of $tau$-trainability and show that current plasticity preserving algorithms can be unified under this framework.<n> Experiments on continual supervised and reinforcement learning tasks confirm that combining these two regularizers effectively preserves plasticity.
  arXiv  Detail & Related papers  (2025-09-26T13:28:53Z)
• Task-Agnostic Federated Continual Learning via Replay-Free Gradient Projection [23.99793728516052]
  Federated continual learning (FCL) enables distributed client devices to learn from streaming data across diverse and evolving tasks.<n>We propose Federated gradient Projection-based Continual Learning with Task Identity Prediction (FedProTIP)<n>This framework mitigates forgetting by projecting client updates onto the complement of the subspace spanned by previously learned representations of the global model.
  arXiv  Detail & Related papers  (2025-09-25T21:20:00Z)
• Learning without Isolation: Pathway Protection for Continual Learning [64.3476595369537]
  Deep networks are prone to catastrophic forgetting during sequential task learning.<n>We propose a novel CL framework, learning without isolation(LwI), where model fusion is formulated as graph matching.<n>Thanks to the sparsity of activation channels in a deep network, LwI can adaptively allocate available pathways for a new task, realizing pathway protection and addressing catastrophic forgetting.
  arXiv  Detail & Related papers  (2025-05-24T07:16:55Z)
• Continual Task Learning through Adaptive Policy Self-Composition [54.95680427960524]
  CompoFormer is a structure-based continual transformer model that adaptively composes previous policies via a meta-policy network. Our experiments reveal that CompoFormer outperforms conventional continual learning (CL) methods, particularly in longer task sequences.
  arXiv  Detail & Related papers  (2024-11-18T08:20:21Z)
• Continual Diffuser (CoD): Mastering Continual Offline Reinforcement Learning with Experience Rehearsal [54.93261535899478]
  In real-world applications, such as robotic control of reinforcement learning, the tasks are changing, and new tasks arise in a sequential order.<n>This situation poses the new challenge of plasticity-stability trade-off for training an agent who can adapt to task changes and retain acquired knowledge.<n>We propose a rehearsal-based continual diffusion model, called Continual diffuser (CoD), to endow the diffuser with the capabilities of quick adaptation (plasticity) and lasting retention (stability)
  arXiv  Detail & Related papers  (2024-09-04T08:21:47Z)
• Towards Robust Continual Learning with Bayesian Adaptive Moment Regularization [51.34904967046097]
  Continual learning seeks to overcome the challenge of catastrophic forgetting, where a model forgets previously learnt information. We introduce a novel prior-based method that better constrains parameter growth, reducing catastrophic forgetting. Results show that BAdam achieves state-of-the-art performance for prior-based methods on challenging single-headed class-incremental experiments.
  arXiv  Detail & Related papers  (2023-09-15T17:10:51Z)
• Task-Adaptive Saliency Guidance for Exemplar-free Class Incremental Learning [60.501201259732625]
  We introduce task-adaptive saliency for EFCIL and propose a new framework, which we call Task-Adaptive Saliency Supervision (TASS) Our experiments demonstrate that our method can better preserve saliency maps across tasks and achieve state-of-the-art results on the CIFAR-100, Tiny-ImageNet, and ImageNet-Subset EFCIL benchmarks.
  arXiv  Detail & Related papers  (2022-12-16T02:43:52Z)
• Consistency is the key to further mitigating catastrophic forgetting in continual learning [14.674494335647841]
  Experience Replay (ER) does not perform well under low-buffer regimes and longer task sequences. consistency in predictions of soft-targets can assist ER in preserving information pertaining to previous tasks better. We propose to cast consistency regularization as a self-supervised pretext task.
  arXiv  Detail & Related papers  (2022-07-11T16:44:49Z)
• Unsupervised Continual Learning via Self-Adaptive Deep Clustering Approach [20.628084936538055]
  Knowledge Retention in Self-Adaptive Deep Continual Learner, (KIERA) is proposed in this paper. KIERA is developed from the notion of flexible deep clustering approach possessing an elastic network structure to cope with changing environments in the timely manner.
  arXiv  Detail & Related papers  (2021-06-28T10:37:14Z)
• Iterative Network Pruning with Uncertainty Regularization for Lifelong Sentiment Classification [25.13885692629219]
  Lifelong learning is non-trivial for deep neural networks. We propose a novel iterative network pruning with uncertainty regularization method for lifelong sentiment classification.
  arXiv  Detail & Related papers  (2021-06-21T15:34:13Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.