In online continual learning (CL), models trained on changing distributions easily forget previously learned knowledge and bias toward newly received tasks. To address this issue, we present Continual Bias Adaptor (CBA), a bi-level framework that augments the classification network to adapt to catastrophic distribution shifts during training, enabling the network to achieve a stable consolidation of all seen tasks. However, the CBA module adjusts distribution shifts in a class-specific manner, exacerbating the stability gap issue and, to some extent, fails to meet the need for continual testing in online CL. To mitigate this challenge, we further propose a novel class-agnostic CBA module that separately aggregates the posterior probabilities of classes from new and old tasks, and applies a stable adjustment to the resulting posterior probabilities. We combine the two kinds of CBA modules into a unified Dual-CBA module, which thus is capable of adapting to catastrophic distribution shifts and simultaneously meets the real-time testing requirements of online CL. Besides, we propose Incremental Batch Normalization (IBN), a tailored BN module to re-estimate its population statistics for alleviating the feature bias arising from the inner loop optimization problem of our bi-level framework. To validate the effectiveness of the proposed method, we theoretically provide some insights into how it mitigates catastrophic distribution shifts, and empirically demonstrate its superiority through extensive experiments based on four rehearsal-based baselines and three public continual learning benchmarks.
PHEVA, a Privacy-preserving Human-centric Ethical Video Anomaly detection dataset. By removing pixel information and providing only de-identified human annotations, PHEVA safeguards personally identifiable information. The dataset includes seven indoor/outdoor scenes, featuring one novel, context-specific camera, and offers over 5x the pose-annotated frames compared to the largest previous dataset. This study benchmarks state-of-the-art methods on PHEVA using a comprehensive set of metrics, including the 10% Error Rate (10ER), a metric used for anomaly detection for the first time providing insights relevant to real-world deployment. As the first of its kind, PHEVA bridges the gap between conventional training and real-world deployment by introducing continual learning benchmarks, with models outperforming traditional methods in 82.14% of cases. The dataset is publicly available at https://github.com/TeCSAR-UNCC/PHEVA.git.
Continual learning and machine unlearning are crucial challenges in machine learning, typically addressed separately. Continual learning focuses on adapting to new knowledge while preserving past information, whereas unlearning involves selectively forgetting specific subsets of data. In this paper, we introduce a new framework that jointly tackles both tasks by leveraging controlled knowledge distillation. Our approach enables efficient learning with minimal forgetting and effective targeted unlearning. By incorporating a fixed memory buffer, the system supports learning new concepts while retaining prior knowledge. The distillation process is carefully managed to ensure a balance between acquiring new information and forgetting specific data as needed. Experimental results on benchmark datasets show that our method matches or exceeds the performance of existing approaches in both continual learning and machine unlearning. This unified framework is the first to address both challenges simultaneously, paving the way for adaptable models capable of dynamic learning and forgetting while maintaining strong overall performance. Source code: \textcolor{blue}{https://respailab.github.io/CLMUL}
Large vision-language models (VLMs) have shown significant performance boost in various application domains. However, adopting them to deal with several sequentially encountered tasks has been challenging because finetuning a VLM on a task normally leads to reducing its generalization power and the capacity of learning new tasks as well as causing catastrophic forgetting on previously learned tasks. Enabling using VLMs in multimodal continual learning (CL) settings can help to address such scenarios. To improve generalization capacity and prevent catastrophic forgetting, we propose a novel prompt-based CL method for VLMs, namely $\textbf{Clu}$ster-based $\textbf{Mo}$dality Fusion Prompt (\textbf{CluMo}). We design a novel \textbf{Key-Key-Prompt} pair, where each prompt is associated with a visual prompt key and a textual prompt key. We adopt a two-stage training strategy. During the first stage, the single-modal keys are trained via $K$-means clustering algorithm to help select the best semantically matched prompt. During the second stage, the prompt keys are frozen, the selected prompt is attached to the input for training the VLM in the CL scenario. Experiments on two benchmarks demonstrate that our method achieves SOTA performance.
Continual learning enables AI models to learn new data sequentially without retraining in real-world scenarios. Most existing methods assume the training data are balanced, aiming to reduce the catastrophic forgetting problem that models tend to forget previously generated data. However, data imbalance and the mixture of new and old data in real-world scenarios lead the model to ignore categories with fewer training samples. To solve this problem, we propose an analytic imbalance rectifier algorithm (AIR), a novel online exemplar-free continual learning method with an analytic (i.e., closed-form) solution for data-imbalanced class-incremental learning (CIL) and generalized CIL scenarios in real-world continual learning. AIR introduces an analytic re-weighting module (ARM) that calculates a re-weighting factor for each class for the loss function to balance the contribution of each category to the overall loss and solve the problem of imbalanced training data. AIR uses the least squares technique to give a non-discriminatory optimal classifier and its iterative update method in continual learning. Experimental results on multiple datasets show that AIR significantly outperforms existing methods in long-tailed and generalized CIL scenarios. The source code is available at https://github.com/fang-d/AIR.
In continual learning (CL), model growth enhances adaptability over new data, improving knowledge retention for more tasks. However, improper model growth can lead to severe degradation of previously learned knowledge, an issue we name as growth-induced forgetting (GIFt), especially in task-agnostic CL using entire grown model for inference. Existing works, despite adopting model growth and random initialization for better adaptability, often fail to recognize the presence of GIFt caused by improper model growth. This oversight limits comprehensive control of forgetting and hinders full utilization of model growth. We are the first in CL to identify this issue and conduct an in-depth study on root cause of GIFt, where layer expansion stands out among model growth strategies, widening layers without affecting model functionality. Yet, direct adoption of layer expansion presents challenges. It lacks data-driven control and initialization of expanded parameters to balance adaptability and knowledge retention. This paper presents a novel SparseGrow approach to overcome the issue of GIFt while enhancing adaptability over new data. SparseGrow employs data-driven sparse layer expansion to control efficient parameter usage during growth, reducing GIFt from excessive growth and functionality changes. It also combines sparse growth with on-data initialization at training late-stage to create partially 0-valued expansions that fit learned distribution, enhancing retention and adaptability. To further minimize forgetting, freezing is applied by calculating the sparse mask, allowing data-driven preservation of important parameters. Through experiments across datasets with various settings, cases, and task numbers, we demonstrate the necessity of layer expansion and showcase the effectiveness of SparseGrow in overcoming GIFt, highlighting its adaptability and knowledge retention for incremental tasks.
An ideal dialogue system requires continuous skill acquisition and adaptation to new tasks while retaining prior knowledge. Dialogue State Tracking (DST), vital in these systems, often involves learning new services and confronting catastrophic forgetting, along with a critical capability loss termed the "Value Selection Quandary." To address these challenges, we introduce the Reason-of-Select (RoS) distillation method by enhancing smaller models with a novel 'meta-reasoning' capability. Meta-reasoning employs an enhanced multi-domain perspective, combining fragments of meta-knowledge from domain-specific dialogues during continual learning. This transcends traditional single-perspective reasoning. The domain bootstrapping process enhances the model's ability to dissect intricate dialogues from multiple possible values. Its domain-agnostic property aligns data distribution across different domains, effectively mitigating forgetting. Additionally, two novel improvements, "multi-value resolution" strategy and Semantic Contrastive Reasoning Selection method, significantly enhance RoS by generating DST-specific selection chains and mitigating hallucinations in teachers' reasoning, ensuring effective and reliable knowledge transfer. Extensive experiments validate the exceptional performance and robust generalization capabilities of our method. The source code is provided for reproducibility.
AA is the process of attributing an unidentified document to its true author from a predefined group of known candidates, each possessing multiple samples. The nature of AA necessitates accommodating emerging new authors, as each individual must be considered unique. This uniqueness can be attributed to various factors, including their stylistic preferences, areas of expertise, gender, cultural background, and other personal characteristics that influence their writing. These diverse attributes contribute to the distinctiveness of each author, making it essential for AA systems to recognize and account for these variations. However, current AA benchmarks commonly overlook this uniqueness and frame the problem as a closed-world classification, assuming a fixed number of authors throughout the system's lifespan and neglecting the inclusion of emerging new authors. This oversight renders the majority of existing approaches ineffective for real-world applications of AA, where continuous learning is essential. These inefficiencies manifest as current models either resist learning new authors or experience catastrophic forgetting, where the introduction of new data causes the models to lose previously acquired knowledge. To address these inefficiencies, we propose redefining AA as CIL, where new authors are introduced incrementally after the initial training phase, allowing the system to adapt and learn continuously. To achieve this, we briefly examine subsequent CIL approaches introduced in other domains. Moreover, we have adopted several well-known CIL methods, along with an examination of their strengths and weaknesses in the context of AA. Additionally, we outline potential future directions for advancing CIL AA systems. As a result, our paper can serve as a starting point for evolving AA systems from closed-world models to continual learning through CIL paradigms.
A continual learning agent builds on previous experiences to develop increasingly complex behaviors by adapting to non-stationary and dynamic environments while preserving previously acquired knowledge. However, scaling these systems presents significant challenges, particularly in balancing the preservation of previous policies with the adaptation of new ones to current environments. This balance, known as the stability-plasticity dilemma, is especially pronounced in complex multi-agent domains such as the train scheduling problem, where environmental and agent behaviors are constantly changing, and the search space is vast. In this work, we propose addressing these challenges in the train scheduling problem using curriculum learning. We design a curriculum with adjacent skills that build on each other to improve generalization performance. Introducing a curriculum with distinct tasks introduces non-stationarity, which we address by proposing a new algorithm: Continual Deep Q-Network (DQN) Expansion (CDE). Our approach dynamically generates and adjusts Q-function subspaces to handle environmental changes and task requirements. CDE mitigates catastrophic forgetting through EWC while ensuring high plasticity using adaptive rational activation functions. Experimental results demonstrate significant improvements in learning efficiency and adaptability compared to RL baselines and other adapted methods for continual learning, highlighting the potential of our method in managing the stability-plasticity dilemma in the adaptive train scheduling setting.
Despite recent progress in enhancing the efficacy of Open-Domain Continual Learning (ODCL) in Vision-Language Models (VLM), failing to (1) correctly identify the Task-ID of a test image and (2) use only the category set corresponding to the Task-ID, while preserving the knowledge related to each domain, cannot address the two primary challenges of ODCL: forgetting old knowledge and maintaining zero-shot capabilities, as well as the confusions caused by category-relatedness between domains. In this paper, we propose a simple yet effective solution: leveraging intra-domain category-aware prototypes for ODCL in CLIP (DPeCLIP), where the prototype is the key to bridging the above two processes. Concretely, we propose a training-free Task-ID discriminator method, by utilizing prototypes as classifiers for identifying Task-IDs. Furthermore, to maintain the knowledge corresponding to each domain, we incorporate intra-domain category-aware prototypes as domain prior prompts into the training process. Extensive experiments conducted on 11 different datasets demonstrate the effectiveness of our approach, achieving 2.37% and 1.14% average improvement in class-incremental and task-incremental settings, respectively.
Continual learning has emerged as a crucial paradigm for learning from sequential data while preserving previous knowledge. In the realm of continual graph learning, where graphs continuously evolve based on streaming graph data, continual graph learning presents unique challenges that require adaptive and efficient graph learning methods in addition to the problem of catastrophic forgetting. The first challenge arises from the interdependencies between different graph data, where previous graphs can influence new data distributions. The second challenge lies in the efficiency concern when dealing with large graphs. To addresses these two problems, we produce an Efficient Continual Graph Learner (E-CGL) in this paper. We tackle the interdependencies issue by demonstrating the effectiveness of replay strategies and introducing a combined sampling strategy that considers both node importance and diversity. To overcome the limitation of efficiency, E-CGL leverages a simple yet effective MLP model that shares weights with a GCN during training, achieving acceleration by circumventing the computationally expensive message passing process. Our method comprehensively surpasses nine baselines on four graph continual learning datasets under two settings, meanwhile E-CGL largely reduces the catastrophic forgetting problem down to an average of -1.1%. Additionally, E-CGL achieves an average of 15.83x training time acceleration and 4.89x inference time acceleration across the four datasets. These results indicate that E-CGL not only effectively manages the correlation between different graph data during continual training but also enhances the efficiency of continual learning on large graphs. The code is publicly available at https://github.com/aubreygjh/E-CGL.
Parameter-efficient fine-tuning (PEFT) methods are increasingly used with pre-trained language models (PLMs) for continual learning (CL). These methods typically involve training a PEFT module for each new task and employing similarity-based selection to route modules during inference. However, they face two major limitations: 1) interference during module training with already learned modules and 2) suboptimal routing when composing modules. In this paper, we present L2R, a method that isolates the training of new PEFT modules to ensure their task specialization. L2R then learns to compose the learned modules by training a network of routers that leverages a small memory containing examples of previously seen tasks. We evaluate our method in two CL setups using various benchmarks. Our results demonstrate that L2R provides an effective composition of PEFT modules, leading to improved generalization and performance compared to other methods.
In recent years, continual learning with pre-training (CLPT) has received widespread interest, instead of its traditional focus of training from scratch. The use of strong pre-trained models (PTMs) can greatly facilitate knowledge transfer and alleviate catastrophic forgetting, but also suffers from progressive overfitting of pre-trained knowledge into specific downstream tasks. A majority of current efforts often keep the PTMs frozen and incorporate task-specific prompts to instruct representation learning, coupled with a prompt selection process for inference. However, due to the limited capacity of prompt parameters, this strategy demonstrates only sub-optimal performance in continual learning. In comparison, tuning all parameters of PTMs often provides the greatest potential for representation learning, making sequential fine-tuning (Seq FT) a fundamental baseline that has been overlooked in CLPT. To this end, we present an in-depth analysis of the progressive overfitting problem from the lens of Seq FT. Considering that the overly fast representation learning and the biased classification layer constitute this particular problem, we introduce the advanced Slow Learner with Classifier Alignment (SLCA++) framework to unleash the power of Seq FT, serving as a strong baseline approach for CLPT. Our approach involves a Slow Learner to selectively reduce the learning rate of backbone parameters, and a Classifier Alignment to align the disjoint classification layers in a post-hoc fashion. We further enhance the efficacy of SL with a symmetric cross-entropy loss, as well as employ a parameter-efficient strategy to implement Seq FT with SLCA++. Across a variety of continual learning scenarios on image classification benchmarks, our approach provides substantial improvements and outperforms state-of-the-art methods by a large margin. Code: https://github.com/GengDavid/SLCA.
Many machine learning models require setting a parameter that controls their size before training, e.g.~number of neurons in DNNs, or inducing points in GPs. Increasing capacity typically improves performance until all the information from the dataset is captured. After this point, computational cost keeps increasing without improved performance. This leads to the question ``How big is big enough?'' We investigate this problem for Gaussian processes (single-layer neural networks) in continual learning. Here, data becomes available incrementally, and the final dataset size will therefore not be known before training, preventing the use of heuristics for setting a fixed model size. We develop a method to automatically adjust model size while maintaining near-optimal performance. Our experimental procedure follows the constraint that any hyperparameters must be set without seeing dataset properties. For our method, a single hyperparameter setting works well across diverse datasets, showing that it requires less tuning compared to others.
Model merging is an efficient empowerment technique in the machine learning community that does not require the collection of raw training data and does not require expensive computation. As model merging becomes increasingly prevalent across various fields, it is crucial to understand the available model merging techniques comprehensively. However, there is a significant gap in the literature regarding a systematic and thorough review of these techniques. This survey provides a comprehensive overview of model merging methods and theories, their applications in various domains and settings, and future research directions. Specifically, we first propose a new taxonomic approach that exhaustively discusses existing model merging methods. Secondly, we discuss the application of model merging techniques in large language models, multimodal large language models, and 10+ machine learning subfields, including continual learning, multi-task learning, few-shot learning, etc. Finally, we highlight the remaining challenges of model merging and discuss future research directions. A comprehensive list of papers about model merging is available at \url{https://github.com/EnnengYang/Awesome-Model-Merging-Methods-Theories-Applications}.
Identifying the causal relations between interested variables plays a pivotal role in representation learning as it provides deep insights into the dataset. Identifiability, as the central theme of this approach, normally hinges on leveraging data from multiple distributions (intervention, distribution shift, time series, etc.). Despite the exciting development in this field, a practical but often overlooked problem is: what if those distribution shifts happen sequentially? In contrast, any intelligence possesses the capacity to abstract and refine learned knowledge sequentially -- lifelong learning. In this paper, with a particular focus on the nonlinear independent component analysis (ICA) framework, we move one step forward toward the question of enabling models to learn meaningful (identifiable) representations in a sequential manner, termed continual causal representation learning. We theoretically demonstrate that model identifiability progresses from a subspace level to a component-wise level as the number of distributions increases. Empirically, we show that our method achieves performance comparable to nonlinear ICA methods trained jointly on multiple offline distributions and, surprisingly, the incoming new distribution does not necessarily benefit the identification of all latent variables.
Fault diagnosis (FD) is essential for maintaining operational safety and minimizing economic losses by detecting system abnormalities. Recently, deep learning (DL)-driven FD methods have gained prominence, offering significant improvements in precision and adaptability through the utilization of extensive datasets and advanced DL models. Modern industrial environments, however, demand FD methods that can handle new fault types, dynamic conditions, large-scale data, and provide real-time responses with minimal prior information. Although online continual learning (OCL) demonstrates potential in addressing these requirements by enabling DL models to continuously learn from streaming data, it faces challenges such as data redundancy, imbalance, and limited labeled data. To overcome these limitations, we propose SRTFD, a scalable real-time fault diagnosis framework that enhances OCL with three critical methods: Retrospect Coreset Selection (RCS), which selects the most relevant data to reduce redundant training and improve efficiency; Global Balance Technique (GBT), which ensures balanced coreset selection and robust model performance; and Confidence and Uncertainty-driven Pseudo-label Learning (CUPL), which updates the model using unlabeled data for continuous adaptation. Extensive experiments on a real-world dataset and two public simulated datasets demonstrate SRTFD's effectiveness and potential for providing advanced, scalable, and precise fault diagnosis in modern industrial systems.
Language model continual learning (CL) has recently attracted significant interest for its ability to adapt large language models (LLMs) to dynamic real-world scenarios without retraining. A major challenge in this domain is catastrophic forgetting, where models lose previously acquired knowledge upon learning new tasks. Existing approaches commonly utilize multiple parameter-efficient fine-tuning (PEFT) blocks to acquire task-specific knowledge, yet these methods are inefficient and fail to leverage potential knowledge transfer across tasks. In this paper, we introduce a novel CL framework for language models, named Knowledge Identification and Fusion (KIF), which boosts knowledge transfer without depending on memory replay. KIF initially segregates the model into 'skill units' based on parameter dependencies, allowing for more precise control. Subsequently, it employs a novel group-wise knowledge identification technique to ascertain the importance distribution of skill units for a new task. By comparing this importance distribution with those from previous tasks, we implement a fine-grained knowledge fusion strategy that retains task-specific knowledge, thereby preventing forgetting, and updates task-shared knowledge, which facilitates bi-directional knowledge transfer. As a result, KIF achieves an optimal balance between retaining prior knowledge and excelling in new tasks. KIF also demonstrates strong generalizability, making it suitable for various base models and adaptable to PEFT methods like LoRA. Furthermore, it offers notable extensibility, supporting enhancements through integration with memory replay techniques. Comprehensive experiments conducted on two CL benchmarks, involving models ranging from 220M to 7B parameters, affirm the effectiveness of KIF and its variants across different settings.
The fact that accurately predicted information can serve as an energy source paves the way for new approaches to autonomous learning. The energy derived from a sequence of successful predictions can be recycled as an immediate incentive and resource, driving the enhancement of predictive capabilities in AI agents. We propose that, through a series of straightforward meta-architectural adjustments, any unsupervised learning apparatus could achieve complete independence from external energy sources, evolving into a self-sustaining physical system with a strong intrinsic 'drive' for continual learning. This concept, while still purely theoretical, is exemplified through the autoencoder, a quintessential model for unsupervised efficient coding. We use this model to demonstrate how progressive paradigm shifts can profoundly alter our comprehension of learning and intelligence. By reconceptualizing learning as an energy-seeking process, we highlight the potential for achieving true autonomy in learning systems, thereby bridging the gap between algorithmic concepts and physical models of intelligence.
Continual learning (CL) refers to the ability to continually learn over time by accommodating new knowledge while retaining previously learned experience. While this concept is inherent in human learning, current machine learning methods are highly prone to overwrite previously learned patterns and thus forget past experience. Instead, model parameters should be updated selectively and carefully, avoiding unnecessary forgetting while optimally leveraging previously learned patterns to accelerate future learning. Since hand-crafting effective update mechanisms is difficult, we propose meta-learning a transformer-based optimizer to enhance CL. This meta-learned optimizer uses attention to learn the complex relationships between model parameters across a stream of tasks, and is designed to generate effective weight updates for the current task while preventing catastrophic forgetting on previously encountered tasks. Evaluations on benchmark datasets like SplitMNIST, RotatedMNIST, and SplitCIFAR-100 affirm the efficacy of the proposed approach in terms of both forward and backward transfer, even on small sets of labeled data, highlighting the advantages of integrating a meta-learned optimizer within the continual learning framework.