Replay-Based Continual Learning

Recent work applies Large Language Models (LLMs) to source-code vulnerability detection, but most evaluations still rely on random train-test splits that ignore time and overestimate real-world performance. In practice, detectors are deployed on evolving code bases and must recognise future vulnerabilities under temporal distribution shift. This paper investigates continual fine-tuning of a decoder-style language model (microsoft/phi-2 with LoRA) on a CVE-linked dataset spanning 2018-2024, organised into bi-monthly windows. We evaluate eight continual learning strategies, including window-only and cumulative training, replay-based baselines and regularisation-based variants. We propose Hybrid Class-Aware Selective Replay (Hybrid-CASR), a confidence-aware replay method for binary vulnerability classification that prioritises uncertain samples while maintaining a balanced ratio of VULNERABLE and FIXED functions in the replay buffer. On bi-monthly forward evaluation Hybrid-CASR achieves a Macro-F1 of 0.667, improving on the window-only baseline (0.651) by 0.016 with statistically significant gains ($p = 0.026$) and stronger backward retention (IBR@1 of 0.741). Hybrid-CASR also reduces training time per window by about 17 percent compared to the baseline, whereas cumulative training delivers only a minor F1 increase (0.661) at a 15.9-fold computational cost. Overall, the results show that selective replay with class balancing offers a practical accuracy-efficiency trade-off for LLM-based temporal vulnerability detection under continuous temporal drift.

Rapid progress in vision-language modeling has enabled pathology report generation from gigapixel whole-slide images, but most approaches assume static training with simultaneous access to all data. In clinical deployment, however, new organs, institutions, and reporting conventions emerge over time, and sequential fine-tuning can cause catastrophic forgetting. We introduce an exemplar-free continual learning framework for WSI-to-report generation that avoids storing raw slides or patch exemplars. The core idea is a compact domain footprint built in a frozen patch-embedding space: a small codebook of representative morphology tokens together with slide-level co-occurrence summaries and lightweight patch-count priors. These footprints support generative replay by synthesizing pseudo-WSI representations that reflect domain-specific morphological mixtures, while a teacher snapshot provides pseudo-reports to supervise the updated model without retaining past data. To address shifting reporting conventions, we distill domain-specific linguistic characteristics into a compact style descriptor and use it to steer generation. At inference, the model identifies the most compatible descriptor directly from the slide signal, enabling domain-agnostic setup without requiring explicit domain identifiers. Evaluated across multiple public continual learning benchmarks, our approach outperforms exemplar-free and limited-buffer rehearsal baselines, highlighting footprint-based generative replay as a practical solution for deployment in evolving clinical settings.

Large language models deployed in the wild must adapt to evolving data, user behavior, and task mixtures without erasing previously acquired capabilities. In practice, this remains difficult: sequential updates induce catastrophic forgetting, while many stabilization methods rely on external procedures that are costly, brittle, or difficult to scale. We present TRC$^{2}$ (Thalamically Routed Cortical Columns), a decoder-only architecture that makes continual learning a property of the backbone itself. TRC$^{2}$ combines stacked cortical columns with a thalamic modulatory pathway for selective inter-column communication and a hippocampal pathway for event selective retrieval, delayed surprise-based writing, and replay-driven consolidation. This design localizes fast plasticity while preserving a slower stable computation pathway. We further introduce a causal memory-update scheme and an online replay controller that adjusts consolidation strength from measured forgetting. Across a task-sequential language-modeling stream over C4, WikiText-103, and GSM8K, TRC$^{2}$ consistently improves task-boundary modeling quality and substantially reduces cumulative forgetting relative to Transformer, Mamba, MoE, DeepSeek and continual learning baselines trained under the same pipeline. Ablations show that the thalamic and hippocampal components are central to the retention gains, while the full model remains competitive in throughput and training cost.

Rehearsal is one of the key techniques for mitigating catastrophic forgetting and has been widely adopted in continual learning algorithms due to its simplicity and practicality. However, the theoretical understanding of how rehearsal scale influences learning dynamics remains limited. To address this gap, we formulate rehearsal-based continual learning as a multidimensional effectiveness-driven iterative optimization problem, providing a unified characterization across diverse performance metrics. Within this framework, we derive a closed-form analysis of adaptability, memorability, and generalization from the perspective of rehearsal scale. Our results uncover several intriguing and counterintuitive findings. First, rehearsal can impair model's adaptability, in sharp contrast to its traditionally recognized benefits. Second, increasing the rehearsal scale does not necessarily improve memory retention. When tasks are similar and noise levels are low, the memory error exhibits a diminishing lower bound. Finally, we validate these insights through numerical simulations and extended analyses on deep neural networks across multiple real-world datasets, revealing statistical patterns of rehearsal mechanisms in continual learning.

Deploying object detection on microcontrollers (MCUs) enables intelligent edge devices but current models cannot learn new object categories after deployment. Existing continual learning methods require storing raw images far exceeding MCU memory budgets of tens of kilobytes. We present Latent Replay Detection (LRD), the first framework for continual object detection under MCU memory constraints. Our key contributions are: 1. Task-Adaptive Compression: Unlike fixed PCA, we propose learnable compression with FiLM (Feature-wise Linear Modulation) conditioning, where task specific embeddings modulate the compression to preserve discriminative features for each task's distribution; 2. Spatial-Diverse Exemplar Selection: Traditional sampling ignores spatial information critical for detection - we select exemplars maximizing bounding box diversity via farthest-point sampling in IoU space, preventing localization bias in replay; 3. MCU-Deployable System: Our latent replay stores 150 bytes per sample versus >10KB for images, enabling a 64KB buffer to hold 400+ exemplars. Experiments on CORe50 (50 classes, 5 tasks) demonstrate that LRD achieves mAP@50 on the initial task and maintains strong performance across subsequent tasks - a significant improvement over naive fine-tuning while operating within strict MCU constraints. Our task-adaptive FiLM compression and spatial diverse exemplar selection work synergistically to preserve detection capabilities. Deployed on STM32H753ZI, ESP32-S3, and MAX78000 MCUs, LRD achieves 4.9-97.5ms latency per inference within a 64KB memory budget-enabling practical continual detection on edge devices for the first time.

Deploying continual object detection on microcontrollers (MCUs) with under 100KB memory requires efficient feature compression that can adapt to evolving task distributions. Existing approaches rely on fixed compression strategies (e.g., FiLM conditioning) that cannot adapt to heterogeneous task characteristics, leading to suboptimal memory utilization and catastrophic forgetting. We introduce Adaptive Hierarchical Compression (AHC), a meta-learning framework featuring three key innovations: (1) true MAML-based compression that adapts via gradient descent to each new task in just 5 inner-loop steps, (2) hierarchical multi-scale compression with scale-aware ratios (8:1 for P3, 6.4:1 for P4, 4:1 for P5) matching FPN redundancy patterns, and (3) a dual-memory architecture combining short-term and long-term banks with importance-based consolidation under a hard 100KB budget. We provide formal theoretical guarantees bounding catastrophic forgetting as O(ε{sq.root(T)} + 1/{sq.root(M)}) where ε is compression error, T is task count, and M is memory size. Experiments on CORe50, TiROD, and PASCAL VOC benchmarks with three standard baselines (Fine-tuning,EWC, iCaRL) demonstrate that AHC enables practical continual detection within a 100KB replay budget, achieving competitive accuracy through mean-pooled compressed feature replay combined with EWC regularization and feature distillation.

Continual learning aims to acquire tasks sequentially without catastrophic forgetting, yet standard strategies face a core tradeoff: regularization-based methods (e.g., EWC) can overconstrain updates when task optima are weakly overlapping, while replay-based methods can retain performance but drift due to imperfect replay. We study a hybrid perspective: trust region continual learning that combines generative replay with a Fisher-metric trust region constraint. We show that, under local approximations, the resulting update admits a Model-Agnostic Meta-Learning (MAML) interpretation with a single implicit inner step: replay supplies an old-task gradient signal (query-like), while the Fisher-weighted penalty provides efficient offline curvature shaping (support-like). This yields an emergent meta-learning property in continual learning: the model becomes an initialization that rapidly re-converges to prior task optima after each task transition, without explicitly optimizing a bilevel objective that couples inner adaptation with outer meta-updates. Empirically, on task-incremental diffusion image generation and continual diffusion-policy control, trust region continual learning achieves the best final performance and retention, and accelerates early-task recovery compared with EWC, replay, and continual meta-learning baselines.

Federated Continual Learning (FCL) leverages inter-client collaboration to better balance new knowledge acquisition and old knowledge retention on non-stationary data. However, existing FCL methods struggle to adapt to streaming scenarios where sequential and ephemerally accessible data chunks lack task identifiers and exhibit arbitrary class overlap, leading to confusion between old and new knowledge and an inability to sustain local inference on all encountered classes. To address this, we propose FedKACE with three components: 1) an adaptive mechanism that determines when to switch the inference model from the local to the global one to improve client-side inference performance; 2) a responsive gradient-balanced replay scheme that utilizes the ratio of the squared L2 gradient norms to balance client-specific knowledge between new acquisition and old retention; 3) a holistic buffer maintenance strategy that preserves highly informative and boundary-significant samples to enhance knowledge retention under class overlap.Experiments across multiple scenarios and theoretical analysis demonstrate the effectiveness of FedKACE.

Catastrophic forgetting poses a fundamental challenge in continual learning, particularly when models are quantized for deployment efficiency. We systematically investigate the interplay between quantization precision (FP16, INT8, INT4) and replay buffer strategies in large language models, revealing unexpected dynamics. While FP16 achieves superior initial task performance (74.44% on NLU), we observe a striking inversion on subsequent tasks: quantized models outperform FP16 by 8-15% on final task forward accuracy, with INT4 achieving nearly double FP16's performance on Code generation (40% vs 20%). Critically, even minimal replay buffers (0.1%) dramatically improve retention - increasing NLU retention after Math training from 45% to 65% across all precision levels - with INT8 consistently achieving the optimal balance between learning plasticity and knowledge retention. We hypothesize that quantization-induced noise acts as implicit regularization, preventing the overfitting to new task gradients that plagues high-precision models. These findings challenge the conventional wisdom that higher precision is always preferable, suggesting instead that INT8 quantization offers both computational efficiency and superior continual learning dynamics. Our results provide practical guidelines for deploying compressed models in continual learning scenarios: small replay buffers (1-2%) suffice for NLU tasks, while Math and Code benefit from moderate buffers (5-10%), with quantized models requiring less replay than FP16 to achieve comparable retention. Code is available at https://github.com/Festyve/LessIsMore.

Data-driven fault diagnosis models for smart grids are usually trained once on a fixed dataset, whereas in operation new fault types appear and monitoring is extended to new grid zones. Retraining from scratch on all accumulated data is costly, while naively updating the model on new data causes catastrophic forgetting. To address this problem, we formulate fault type classification and fault zone localization as continual learning (CL) problems and design four evaluation scenarios on the IEEE 13-node test feeder, three class-incremental and one domain-incremental. We then propose Prototype-based Dark Experience Replay (ProDER), which extends DER++ with prototype attraction and prototype-level repulsion losses that stabilize the feature space, temperature-scaled logit distillation, and a prototype-aware replay memory that retains both core and boundary samples of each class. ProDER achieves the highest accuracy among the tested CL methods in all scenarios, with an average accuracy of 58.2%, 6.6 points above the strongest competing method (DPDMR, 51.6%) and only 3.2 points below joint training (61.4%). Per scenario, it improves over the strongest competitor by 4.2 to 7.4 points and closes the gap to joint training to as little as 1.0 point in fault type classification, while matching it in fault zone localization. Moreover, it remains the best method when the replay buffer is substantially reduced. These results show that prototype-guided replay is an effective, memory-bounded way to keep fault diagnosis models up to date as the grid evolves, while validation on field measurements remains a necessary next step.

Recently, significant progress has been made in multi-modal continual learning, aiming to learn new tasks sequentially in multi-modal settings while preserving performance on previously learned ones. However, existing methods mainly focus on coarse-grained tasks, with limitations in addressing modality entanglement in fine-grained continual learning settings. To bridge this gap, we introduce a novel Continual Audio-Visual Segmentation (CAVS) task, aiming to continuously segment new classes guided by audio. Through comprehensive analysis, two critical challenges are identified: 1) multi-modal semantic drift, where a sounding objects is labeled as background in sequential tasks; 2) co-occurrence confusion, where frequent co-occurring classes tend to be confused. In this work, a Collision-based Multi-modal Rehearsal (CMR) framework is designed to address these challenges. Specifically, for multi-modal semantic drift, a Multi-modal Sample Selection (MSS) strategy is proposed to select samples with high modal consistency for rehearsal. Meanwhile, for co-occurence confusion, a Collision-based Sample Rehearsal (CSR) mechanism is designed, allowing for the increase of rehearsal sample frequency of those confusable classes during training process. Moreover, we construct three audio-visual incremental scenarios to verify effectiveness of our method. Comprehensive experiments demonstrate that our method significantly outperforms single-modal continual learning methods. Code can be seen at https://github.com/cqu-student/CAVS-CMR.

Lifelong Person Re-Identification (LReID) typically resists catastrophic forgetting by replaying historical samples, rehearsing domain distributions, or distilling previous model knowledge. Among these, data replay is favored for its simplicity and efficiency, as it fundamentally relies on storing discrete raw images. Although often claimed to be efficient, repeatedly training on an accumulating replay buffer with complex selection strategies across sequential domains is actually highly inefficient. Furthermore, this discrete selection severely restricts historical data coverage and results in low information density, inevitably leading to poor generalization on evolving domains and causing these methods to gradually fall behind other approaches. In this paper, we rethink LReID replay and shift the paradigm from sample selection to information compression, proposing a High Information Density Replay (HiDeR) framework. Rather than saving sparse instances, we continually consolidate historical data into a compact, fixed-budget memory. Specifically, we introduce a complexity aware allocation mechanism to dynamically assign memory quotas based on intra-class variance, alongside a metric guided condensation objective that directly preserves essential identity topologies. Furthermore, since highly compressed synthetic samples exhibit artifact styles unsuitable for current domain training, we introduce a cross modality adaptation strategy. By bidirectionally translating styles between synthetic and real samples, this strategy bridges the modality gap to mitigate optimization conflicts, while also enriching stylistic diversity for better generalization. Extensive experiments demonstrate that our framework outperforms state-of-the-art methods in retaining historical knowledge and improving overall generalization, while substantially reducing the cumulative replay cost.

Machine learning on heterogeneous graphs has advanced rapidly, driven by the diverse entities, relations, and semantics inherent in real-world data. However, most existing studies assume static graphs, whereas real-world graph data are often continuously updated. Continual learning in this setting is particularly challenging because historical knowledge is encoded not only in target-node representations, but also in their heterogeneous structural contexts and relation-dependent semantic responses. To this end, we introduce HERO, a heterogeneous continual graph learning framework that preserves historical knowledge at both structural and semantic levels while adapting to incoming tasks. We analyze experience replay as an approximation to the unavailable historical gradient and decompose its error into target-selection and context-reconstruction terms. This analysis shows that replaying target nodes alone is generally insufficient for relation-sensitive HGNNs, as removing typed neighborhood context can alter both representations and historical gradients. Motivated by this, HERO employs DiSCo to select representative target nodes and reconstruct compact heterogeneous contexts through relation-aware multi-type neighbor expansion. To further preserve knowledge that cannot be captured by replayed subgraphs alone, HERO aligns task-conditioned predictions and heterogeneous semantic responses through multi-level knowledge distillation. A lightweight look-ahead adaptation step is additionally used to improve plasticity on incoming tasks. Experiments on four datasets with three HGNN backbones show that HERO achieves the best or tied-best average performance among state-of-the-art baselines, while maintaining competitive forgetting. Our code is available at https://github.com/gqBond/HCGL.

Pretrained model-based incremental object detection (PTMIOD) leverages the rich detection priors of pretrained detectors to learn new categories incrementally while preserving detection ability on previously learned ones. Existing methods mainly exploit pretrained detectors as a whole, without explicitly distinguishing which components should remain stable and which require plastic adaptation. In this paper, we revisit PTMIOD from a component-wise stability-plasticity perspective. Our analysis of pretrained DETR-based detectors reveals a clear asymmetry: localization heads preserve transferable geometric priors across tasks, whereas classification-related representations require greater plasticity to handle new categories, especially in cross-domain scenarios where downstream data deviate from the pretraining domain. Based on this finding, we propose a selective adaptation and retention framework that freezes explicit localization heads to preserve localization stability, while adapting transformer representations with parameter-efficient fine-tuning and updating classification heads for classification-oriented plasticity. To alleviate classification-side forgetting, we pioneer the use of pseudo-feature replay in PTMIOD and design Quality-aware Gaussian Feature Replay, which estimates reliable class-wise feature distributions from high-quality matched object features and replays sampled pseudo features to maintain old-class decision boundaries. Since continual adaptation can shift the feature space and undermine replayed distributions, we further develop Two-stage Consistent Distillation to align teacher and student representations at both proposal generation and refinement stages.Extensive experiments on COCO, VOC, and TT100K show that our method achieves state-of-the-art performance, demonstrating a favorable balance between old-class retention and new-class adaptation.