Class-Incremental Learning (CIL) aims to prevent catastrophic forgetting of previously learned classes while sequentially incorporating new ones. The more challenging Few-shot CIL (FSCIL) setting further complicates this by providing only a limited number of samples for each new class, increasing the risk of overfitting in addition to standard CIL challenges. While catastrophic forgetting has been extensively studied, overfitting in FSCIL, especially with large foundation models, has received less attention. To fill this gap, we propose the Singular Value Fine-tuning for FSCIL (SVFCL) and compared it with existing approaches for adapting foundation models to FSCIL, which primarily build on Parameter Efficient Fine-Tuning (PEFT) methods like prompt tuning and Low-Rank Adaptation (LoRA). Specifically, SVFCL applies singular value decomposition to the foundation model weights, keeping the singular vectors fixed while fine-tuning the singular values for each task, and then merging them. This simple yet effective approach not only alleviates the forgetting problem but also mitigates overfitting more effectively while significantly reducing trainable parameters. Extensive experiments on four benchmark datasets, along with visualizations and ablation studies, validate the effectiveness of SVFCL. The code will be made available.
Class-incremental learning (CIL) aims to continuously adapt to emerging classes while retaining knowledge of previously learned ones. Few-shot class-incremental learning (FSCIL) presents an even greater challenge which requires the model to learn incremental classes with only a limited number of samples. In conventional CIL, joint training is widely considered the upper bound, serving as both a benchmark and a methodological guide. However, we find that joint training fails to be a meaningful upper bound in FSCIL due to the inherent difficulty of inter-task class separation (ICS) caused by severe class imbalance. In this work, we introduce a new joint training benchmark tailored for FSCIL by integrating imbalance-aware techniques, effectively bridging the performance gap between base and incremental classes. Furthermore, we point out inconsistencies in the experimental setup and evaluation of existing FSCIL methods. To ensure fair comparisons between different FSCIL approaches and joint training, we standardize training conditions and propose a unified evaluation protocol that simultaneously considers the validation set and computational complexity. By establishing a reliable upper bound and a standardized evaluation framework for FSCIL, our work provides a clear benchmark and a practical foundation for future research.
Class-Incremental Learning (CIL) enables learning systems to continuously adapt to evolving data streams. With the advancement of pre-training, leveraging pre-trained vision-language models (e.g., CLIP) offers a promising starting point for CIL. However, CLIP makes decisions by matching visual embeddings to class names, overlooking the rich contextual information conveyed through language. For instance, the concept of ``cat'' can be decomposed into features like tail, fur, and face for recognition. Besides, since the model is continually updated, these detailed features are overwritten in CIL, requiring external knowledge for compensation. In this paper, we introduce ExterNal knowledGe INjEction (ENGINE) for CLIP-based CIL. To enhance knowledge transfer from outside the dataset, we propose a dual-branch injection tuning framework that encodes informative knowledge from both visual and textual modalities. The visual branch is enhanced with data augmentation to enrich the visual features, while the textual branch leverages GPT-4 to rewrite discriminative descriptors. In addition to this on-the-fly knowledge injection, we also implement post-tuning knowledge by re-ranking the prediction results during inference. With the injected knowledge, the model can better capture informative features for downstream tasks as data evolves. Extensive experiments demonstrate the state-of-the-art performance of ENGINE. Code is available at: https://github.com/RenaissCode/ENGINE
Exemplar-Free Class-Incremental Learning (EFCIL) aims to sequentially learn from distinct categories without retaining exemplars but easily suffers from catastrophic forgetting of learned knowledge. While existing EFCIL methods leverage knowledge distillation to alleviate forgetting, they still face two critical challenges: semantic shift and decision bias. Specifically, the embeddings of old tasks shift in the embedding space after learning new tasks, and the classifier becomes biased towards new tasks due to training solely with new data, hindering the balance between old and new knowledge. To address these issues, we propose the Dual-Projection Shift Estimation and Classifier Reconstruction (DPCR) approach for EFCIL. DPCR effectively estimates semantic shift through a dual-projection, which combines a learnable transformation with a row-space projection to capture both task-wise and category-wise shifts. Furthermore, to mitigate decision bias, DPCR employs ridge regression to reformulate a classifier reconstruction process. This reconstruction exploits previous in covariance and prototype of each class after calibration with estimated shift, thereby reducing decision bias. Extensive experiments demonstrate that, on various datasets, DPCR effectively balances old and new tasks, outperforming state-of-the-art EFCIL methods. Our codes are available at https://github.com/RHe502/ICML25-DPCR.
Incremental learning (IL) aims to overcome catastrophic forgetting of previous tasks while learning new ones. Existing IL methods make strong assumptions that the incoming task type will either only increases new classes or domains (i.e. Class IL, Domain IL), or increase by a static scale in a class- and domain-agnostic manner (i.e. Versatile IL (VIL)), which greatly limit their applicability in the unpredictable and dynamic wild. In this work, we investigate $\textbf{Universal Incremental Learning (UIL)}$, where a model neither knows which new classes or domains will increase along sequential tasks, nor the scale of the increments within each task. This uncertainty prevents the model from confidently learning knowledge from all task distributions and symmetrically focusing on the diverse knowledge within each task distribution. Consequently, UIL presents a more general and realistic IL scenario, making the model face confusion arising from inter-task and intra-task distribution randomness. To $\textbf{Mi}$tigate both $\textbf{Co}$nfusion, we propose a simple yet effective framework for UIL, named $\textbf{MiCo}$. At the inter-task distribution level, we employ a multi-objective learning scheme to enforce accurate and deterministic predictions, and its effectiveness is further enhanced by a direction recalibration module that reduces conflicting gradients. Moreover, at the intra-task distribution level, we introduce a magnitude recalibration module to alleviate asymmetrical optimization towards imbalanced class distribution. Extensive experiments on three benchmarks demonstrate the effectiveness of our method, outperforming existing state-of-the-art methods in both the UIL scenario and the VIL scenario. Our code will be available at $\href{https://github.com/rolsheng/UIL}{here}$.
Class-incremental learning (CIL) for time series data faces critical challenges in balancing stability against catastrophic forgetting and plasticity for new knowledge acquisition, particularly under real-world constraints where historical data access is restricted. While pre-trained models (PTMs) have shown promise in CIL for vision and NLP domains, their potential in time series class-incremental learning (TSCIL) remains underexplored due to the scarcity of large-scale time series pre-trained models. Prompted by the recent emergence of large-scale pre-trained models (PTMs) for time series data, we present the first exploration of PTM-based Time Series Class-Incremental Learning (TSCIL). Our approach leverages frozen PTM backbones coupled with incrementally tuning the shared adapter, preserving generalization capabilities while mitigating feature drift through knowledge distillation. Furthermore, we introduce a Feature Drift Compensation Network (DCN), designed with a novel two-stage training strategy to precisely model feature space transformations across incremental tasks. This allows for accurate projection of old class prototypes into the new feature space. By employing DCN-corrected prototypes, we effectively enhance the unified classifier retraining, mitigating model feature drift and alleviating catastrophic forgetting. Extensive experiments on five real-world datasets demonstrate state-of-the-art performance, with our method yielding final accuracy gains of 1.4%-6.1% across all datasets compared to existing PTM-based approaches. Our work establishes a new paradigm for TSCIL, providing insights into stability-plasticity optimization for continual learning systems.
Binary Neural Networks (BNNs) are a promising approach to enable Artificial Neural Network (ANN) implementation on ultra-low power edge devices. Such devices may compute data in highly dynamic environments, in which the classes targeted for inference can evolve or even novel classes may arise, requiring continual learning. Class Incremental Learning (CIL) is a common type of continual learning for classification problems, that has been scarcely addressed in the context of BNNs. Furthermore, most of existing BNNs models are not fully binary, as they require several real-valued network layers, at the input, the output, and for batch normalization. This paper goes a step further, enabling class incremental learning in Fully-Binarized NNs (FBNNs) through four main contributions. We firstly revisit the FBNN design and its training procedure that is suitable to CIL. Secondly, we explore loss balancing, a method to trade-off the performance of past and current classes. Thirdly, we propose a semi-supervised method to pre-train the feature extractor of the FBNN for transferable representations. Fourthly, two conventional CIL methods, \ie, Latent and Native replay, are thoroughly compared. These contributions are exemplified first on the CIFAR100 dataset, before being scaled up to address the CORE50 continual learning benchmark. The final results based on our 3Mb FBNN on CORE50 exhibit at par and better performance than conventional real-valued larger NN models.
Many studies combine text and audio to capture multi-modal information but they overlook the model's generalization ability on new datasets. Introducing new datasets may affect the feature space of the original dataset, leading to catastrophic forgetting. Meanwhile, large model parameters can significantly impact training performance. To address these limitations, we introduce a novel task called Text-Audio Incremental Learning (TAIL) task for text-audio retrieval, and propose a new method, PTAT, Prompt Tuning for Audio-Text incremental learning. This method utilizes prompt tuning to optimize the model parameters while incorporating an audio-text similarity and feature distillation module to effectively mitigate catastrophic forgetting. We benchmark our method and previous incremental learning methods on AudioCaps, Clotho, BBC Sound Effects and Audioset datasets, and our method outperforms previous methods significantly, particularly demonstrating stronger resistance to forgetting on older datasets. Compared to the full-parameters Finetune (Sequential) method, our model only requires 2.42\% of its parameters, achieving 4.46\% higher performance.
We introduce Frank, a human-in-the-loop system for co-evolutionary hybrid decision-making aiding the user to label records from an un-labeled dataset. Frank employs incremental learning to ``evolve'' in parallel with the user's decisions, by training an interpretable machine learning model on the records labeled by the user. Furthermore, Frank advances state-of-the-art approaches by offering inconsistency controls, explanations, fairness checks, and bad-faith safeguards simultaneously. We evaluate our proposal by simulating the users' behavior with various levels of expertise and reliance on Frank's suggestions. The experiments show that Frank's intervention leads to improvements in the accuracy and the fairness of the decisions.
—Specific emitter identification (SEI) is a non-password authentication method that adds an extra layer of security to wireless devices. However, existing SEI methods are unable to continuously learn new classes from a limited number of training examples due to data scarcity, which is more challenging than the catastrophic forgetting and overfitting problems associated with the widely studied class-incremental learning (CIL). In this article, we propose a novel few-shot class-incremental specific emitter identification (FSCIL-SEI) framework to address the challenge of catastrophic forgetting and overfitting in CIL. Specifically, to ensure interclass discriminability during the incremental process, we first employ prototype learning training methods in the base task and introduce a self-supervised contrastive learning (SSCL) that increases interclass distances and reduces intraclass distances in the feature space. Second, we propose a separation of class weights (SCWs) to isolate old and new class weights in the classification layer, which effectively mitigates the issue of catastrophic forgetting. Finally, to alleviate the problem of overfitting due to insufficient samples during incremental training, we introduce a three-stage course learning (CL) approach that advances from simple to complex tasks, which not only mitigates overfitting but also improves the generalization ability of the model. Experimental results demonstrate that our method outperforms other FSCIL methods in terms of both performance degradation (PD) and incremental accuracy when evaluated on automatic identification system (AIS) and automatic dependent surveillance–broadcast (ADS-B) datasets
The ability to learn new concepts while preserve the learned knowledge is desirable for learning systems in Class-Incremental Learning (CIL). Recently, feature expansion of the model become a prevalent solution for CIL, where the old features are fixed during the training of the new task while new features are expanded for the new tasks. However, such task-specific features learned from the new task may collide with the old features, leading to misclassification between tasks. Therefore, the expanded model is often encouraged to capture diverse features from the new task, aiming to avoid such collision. However, the existing solution is largely restricted to the samples from the current task, because of the poor accessibility to previous samples. To promote the learning and transferring of diverse features across tasks, we propose a framework called Task-Agnostic Guided Feature Expansion (TagFex). Firstly, it captures task-agnostic features continually with a separate model, providing extra task-agnostic features for subsequent tasks. Secondly, to obtain useful features from the task-agnostic model for the current task, it aggregates the task-agnostic features with the task-specific feature using a merge attention. Then the aggregated feature is transferred back into the task-specific feature for inference, helping the task-specific model capture diverse features. Extensive experiments show the effectiveness and superiority of TagFex on various CIL settings. Code is available at https://github.com/bwnzheng/TagFex_CVPR2025.
Current research on class-incremental learning primarily focuses on single-label classification tasks. However, real-world applications often involve multi-label scenarios, such as image retrieval and medical imaging. Therefore, this paper focuses on the challenging yet practical multi-label class-incremental learning (MLCIL) problem. In addition to the challenge of catastrophic forgetting, MLCIL encounters issues related to feature confusion, encompassing inter-session and intra-feature confusion. To address these problems, we propose a novel MLCIL approach called class-independent increment (CLIN). Specifically, in contrast to existing methods that extract image-level features, we propose a class-independent incremental network (CINet) to extract multiple class-level embeddings for multi-label samples. It learns and preserves the knowledge of different classes by constructing class-specific tokens. On this basis, we develop two novel loss functions, optimizing the learning of class-specific tokens and class-level embeddings, respectively. These losses aim to distinguish between new and old classes, further alleviating the problem of feature confusion. Extensive experiments on MS-COCO and PASCAL VOC datasets demonstrate the effectiveness of our method for improving recognition performance and mitigating forgetting on various MLCIL tasks.
Class Incremental Learning (CIL) requires a model to continuously learn new classes without forgetting previously learned ones. While recent studies have significantly alleviated the problem of catastrophic forgetting (CF), more and more research reveals that the order in which classes appear have significant influences on CIL models. Specifically, prioritizing the learning of classes with lower similarity will enhance the model's generalization performance and its ability to mitigate forgetting. Hence, it is imperative to develop an order-robust class incremental learning model that maintains stable performance even when faced with varying levels of class similarity in different orders. In response, we first provide additional theoretical analysis, which reveals that when the similarity among a group of classes is lower, the model demonstrates increased robustness to the class order. Then, we introduce a novel \textbf{G}raph-\textbf{D}riven \textbf{D}ynamic \textbf{S}imilarity \textbf{G}rouping (\textbf{GDDSG}) method, which leverages a graph coloring algorithm for class-based similarity grouping. The proposed approach trains independent CIL models for each group of classes, ultimately combining these models to facilitate joint prediction. Experimental results demonstrate that our method effectively addresses the issue of class order sensitivity while achieving optimal performance in both model accuracy and anti-forgetting capability. Our code is available at https://github.com/AIGNLAI/GDDSG.
Incremental Learning scenarios do not always represent real-world inference use-cases, which tend to have less strict task boundaries, and exhibit repetition of common classes and concepts in their continual data stream. To better represent these use-cases, new scenarios with partial repetition and mixing of tasks are proposed, where the repetition patterns are innate to the scenario and unknown to the strategy. We investigate how exemplar-free incremental learning strategies are affected by data repetition, and we adapt a series of state-of-the-art approaches to analyse and fairly compare them under both settings. Further, we also propose a novel method (Horde), able to dynamically adjust an ensemble of self-reliant feature extractors, and align them by exploiting class repetition. Our proposed exemplar-free method achieves competitive results in the classic scenario without repetition, and state-of-the-art performance in the one with repetition.
Few-shot class-incremental learning (FSCIL) poses significant challenges for artificial neural networks due to the need to efficiently learn from limited data while retaining knowledge of previously learned tasks. Inspired by the brain's mechanisms for categorization and analogical learning, we propose a novel approach called Brain-inspired Analogical Mixture Prototypes (BAMP). BAMP has three components: mixed prototypical feature learning, statistical analogy, and soft voting. Starting from a pre-trained Vision Transformer (ViT), mixed prototypical feature learning represents each class using a mixture of prototypes and fine-tunes these representations during the base session. The statistical analogy calibrates the mean and covariance matrix of prototypes for new classes according to similarity to the base classes, and computes classification score with Mahalanobis distance. Soft voting combines both merits of statistical analogy and an off-shelf FSCIL method. Our experiments on benchmark datasets demonstrate that BAMP outperforms state-of-the-art on both traditional big start FSCIL setting and challenging small start FSCIL setting. The study suggests that brain-inspired analogical mixture prototypes can alleviate catastrophic forgetting and over-fitting problems in FSCIL.
Model-free power flow calculation, driven by the rise of smart meter (SM) data and the lack of network topology, often relies on artificial intelligence neural networks (ANNs). However, training ANNs require vast amounts of SM data, posing privacy risks for households in distribution networks. To ensure customers' privacy during the SM data gathering and online sharing, we introduce a privacy preserving PF calculation framework, composed of two local strategies: a local randomisation strategy (LRS) and a local zero-knowledge proof (ZKP)-based data collection strategy. First, the LRS is used to achieve irreversible transformation and robust privacy protection for active and reactive power data, thereby ensuring that personal data remains confidential. Subsequently, the ZKP-based data collecting strategy is adopted to securely gather the training dataset for the ANN, enabling SMs to interact with the distribution system operator without revealing the actual voltage magnitude. Moreover, to mitigate the accuracy loss induced by the seasonal variations in load profiles, an incremental learning strategy is incorporated into the online application. The results across three datasets with varying measurement errors demonstrate that the proposed framework efficiently collects one month of SM data within one hour. Furthermore, it robustly maintains mean errors of 0.005 p.u. and 0.014 p.u. under multiple measurement errors and seasonal variations in load profiles, respectively.
WiFi-based human activity recognition (HAR) holds significant application potential across various fields. To handle dynamic environments where new activities are continuously introduced, WiFi-based HAR systems must adapt by learning new concepts without forgetting previously learned ones. Furthermore, retaining knowledge from old activities by storing historical exemplar is impractical for WiFi-based HAR due to privacy concerns and limited storage capacity of edge devices. In this work, we propose ConSense, a lightweight and fast-adapted exemplar-free class incremental learning framework for WiFi-based HAR. The framework leverages the transformer architecture and involves dynamic model expansion and selective retraining to preserve previously learned knowledge while integrating new information. Specifically, during incremental sessions, small-scale trainable parameters that are trained specifically on the data of each task are added in the multi-head self-attention layer. In addition, a selective retraining strategy that dynamically adjusts the weights in multilayer perceptron based on the performance stability of neurons across tasks is used. Rather than training the entire model, the proposed strategies of dynamic model expansion and selective retraining reduce the overall computational load while balancing stability on previous tasks and plasticity on new tasks. Evaluation results on three public WiFi datasets demonstrate that ConSense not only outperforms several competitive approaches but also requires fewer parameters, highlighting its practical utility in class-incremental scenarios for HAR.
While 3D instance segmentation (3DIS) has advanced significantly, existing methods typically assume that all object classes are known in advance and are uniformly distributed. However, this assumption is unrealistic in dynamic, real-world environments where new classes emerge gradually and exhibit natural imbalance. Although some approaches have addressed class emergence, they often overlook class imbalance, resulting in suboptimal performance -- particularly on rare categories. To tackle this challenge, we propose CLIMB-3D, a unified framework for \textbf{CL}ass-incremental \textbf{Imb}alance-aware \textbf{3D}IS. Building upon established exemplar replay (ER) strategies, we show that ER alone is insufficient to achieve robust performance under constrained memory conditions. To mitigate this, we introduce a novel pseudo-label generator (PLG) that extends supervision to previously learned categories by leveraging predictions from a frozen prior model. Despite its promise, PLG tends to bias towards frequent classes. Therefore, we propose a class-balanced re-weighting (CBR) scheme, that estimates object frequencies from pseudo-labels and dynamically adjusts training bias -- without requiring access to past data. We design and evaluate three incremental scenarios for 3DIS on the challenging ScanNet200 dataset, and additionally on semantic segmentation on ScanNetV2. Our approach achieves state-of-the-art results, surpassing prior work by up to 16.76\% mAP for instance segmentation and approximately 30\% mIoU for semantic segmentation, demonstrating strong generalization across both frequent and rare classes.
Class-incremental learning requires models to continually acquire knowledge of new classes without forgetting old ones. Although pre-trained models have demonstrated strong performance in class-incremental learning, they remain susceptible to catastrophic forgetting when learning new concepts. Excessive plasticity in the models breaks generalizability and causes forgetting, while strong stability results in insufficient adaptation to new classes. This necessitates effective adaptation with minimal modifications to preserve the general knowledge of pre-trained models. To address this challenge, we first introduce a new parameter-efficient fine-tuning module 'Learn and Calibrate', or LuCA, designed to acquire knowledge through an adapter-calibrator couple, enabling effective adaptation with well-refined feature representations. Second, for each learning session, we deploy a sparse LuCA module on top of the last token just before the classifier, which we refer to as 'Token-level Sparse Calibration and Adaptation', or TOSCA. This strategic design improves the orthogonality between the modules and significantly reduces both training and inference complexity. By leaving the generalization capabilities of the pre-trained models intact and adapting exclusively via the last token, our approach achieves a harmonious balance between stability and plasticity. Extensive experiments demonstrate TOSCA's state-of-the-art performance while introducing ~8 times fewer parameters compared to prior methods.
Current machine learning paradigm relies on continuous representations like neural networks, which iteratively adjust parameters to approximate outcomes rather than directly learning the structure of problem. This spreads information across the network, causing issues like information loss and incomprehensibility Building on prior work in environment dynamics modeling, we propose a method that learns visual space in a structured, continual manner. Our approach refines networks to capture the core structure of objects while representing significant subvariants in structure efficiently. We demonstrate this with 2D shape detection, showing incremental learning on MNIST without overwriting knowledge and creating compact, comprehensible representations. These results offer a promising step toward a transparent, continually learning alternative to traditional neural networks for visual processing.