A Comprehensive Survey of Continual Learning: Theory, Method and Application
Open accessThe DOI matches its Crossref record; 456 of the 456 references it cites were found in a registry or an index.
Abstract
To cope with real-world dynamics, an intelligent system needs to incrementally acquire, update, accumulate, and exploit knowledge throughout its lifetime. This ability, known as continual learning, provides a foundation for AI systems to develop themselves adaptively. In a general sense, continual learning is explicitly limited by catastrophic forgetting, where learning a new task usually results in a dramatic performance drop of the old tasks. Beyond this, increasingly numerous advances have emerged in recent years that largely extend the understanding and application of continual learning. The growing and widespread interest in this direction demonstrates its realistic significance as well as complexity. In this work, we present a comprehensive survey of continual learning, seeking to bridge the basic settings, theoretical foundations, representative methods, and practical applications. Based on existing theoretical and empirical results, we summarize the general objectives of continual learning as ensuring a proper stability-plasticity trade-off and an adequate intra/inter-task generalizability in the context of resource efficiency. Then we provide a state-of-the-art and elaborated taxonomy, extensively analyzing how representative strategies address continual learning, and how they are adapted to particular challenges in various applications. Through an in-depth discussion of promising directions, we believe that such a holistic perspective can greatly facilitate subsequent exploration in this field and beyond.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Complex skills can be synthesized by composing simpler programs, which compounds VOYAGER's capabilities rapidly over time and alleviates catastrophic forgetting in other continual learning methods [31, 32].”
From Voyager: An Open-Ended Embodied Agent with Large Language Models (2023) · cites this paper as [32]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“However, these agents are not lifelong learners that can progressively acquire, update, accumulate, and transfer knowledge over extended time spans [31, 32].”
From Voyager: An Open-Ended Embodied Agent with Large Language Models (2023) · cites this paper as [32]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“This position also makes memory currently the most practical route to continual learning for deployed agents: retraining the underlying model for every user, task, and correction is rarely feasible, so what the system learns after deployment often depends on what it writes down [26, 67].”
From Agentic Memory Management Is the Next Frontier of Information Retrieval (2026) · cites this paper as [67]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“In AI, memory is usually described in terms of persisting state, retaining conversational context, or learning continually without catastrophic forgetting [32, 67].”
From Agentic Memory Management Is the Next Frontier of Information Retrieval (2026) · cites this paper as [67]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“The stability–plasticity dilemma is central to CL, as it balances the model’s ability to retain past knowledge (stability) while acquiring new information (plasticity) [61,62].”
From Deepfake Media Forensics: Status and Future Challenges (2025) · cites this paper as [62]
2 more claims about this paper
Create a free account to see every claim other papers make about this one.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Items 201–400 of 457
Mixed-Privacy Forgetting in Deep Networks
Golatkar, A., Achille, A., Ravichandran, A., Polito, M., & Soatto, S. (2021) · 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr46437.2021.00085 · Cited by 125 · cited 1×
“…oftrainingsamples[307],utilizinghistoricalparametersand gradients[308],removingoldknowledgefromparameterswith Fisherinformationmatrix[309],addingadaptiveparameterson apre-trainedmodel[310],etc. Ontheotherhand,continual learningmaysufferfromdataleakageandprivacyinvasionas retainingalloldknowledge.”
Wang, L., Yang, K., Li, C., Hong, L., Li, Z., & Zhu, J. (2021) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr46437.2021.00534 · Cited by 124 · cited 8×
“Weight regularization [314], [374], [427], [428] and experience replay [157], [428] have been shown to be effective in mitigating catastrophic forgetting of generative models.”
BackgroundMethodInfluentialSelf-citationOn Learning the Geodesic Path for Incremental Learning
Simon, C., Koniusz, P., & Harandi, M. (2021) · 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr46437.2021.00164 · Cited by 121 · cited 3×
“GeoDL [150] performs KD along a path that connects the low-dimensional projections of the old and new feature spaces for a smooth transition between them.”
BackgroundMethodClass-Incremental Learning with Strong Pre-trained Models
Wu, T.-Y., Gurumurthy, S., Li, Z., Ravichandran, A., Vasconcelos, N. M., Bhotika, R., & Soatto, S. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.00938 · Cited by 120 · cited 3×
“In addition to an earlier work on obtaining sparse representations through meta-training [185], recent works have attempted to incorporate the advantages of selfsupervised learning (SSL) [125], [281], [335] and large-scale pre-training [295], [380], [456] to improve the representations in initialization and in continu…”
BackgroundMethodMazumder, P., Singh, P., & Rai, P. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i3.16334 · Cited by 119
Continual Reinforcement Learning with Complex SynapsesPreprint
Kaplanis, C., Shanahan, M., & Clopath, C. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1802.07239 · Cited by 119 · cited 1×
“Similarly, resolving within-task catastrophic forgetting can facilitate reinforcement learning [299], [300] and stabilize the training of GANs [301], [302].”
BackgroundContinual Learning in the Teacher-Student Setup: Impact of Task SimilarityPreprint
Lee, S., Goldt, S., & Saxe, A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2107.04384 · Cited by 118 · cited 2×
“Sometheoreticalstudieswiththeneuraltangent kernel (NTK) [96], [97], [98], [99] suggest that an increase in task similarity may lead to more forgetting.”
BackgroundOnline Continual Learning with Natural Distribution Shifts: An Empirical Study with Visual Data
Cai, Z., Şener, O., & Koltun, V. (2021) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00817 · Cited by 117
Training Networks in Null Space of Feature Covariance for Continual Learning
Wang, S., Li, X., Sun, J., & Xu, Z. (2021) · 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr46437.2021.00025 · Cited by 116
A PAC-Bayesian bound for Lifelong Learning
Pentina, A., & Lampert, C. H. (2013) · arXiv (Cornell University) · doi:10.48550/arxiv.1311.2838 · Cited by 111 · cited 1×
“2(d)), which is further validated by a theoretical analysis of the forgetting-generalization trade-off [94] and the PAC-Bayes bound of generalization errors [69], [95].”
BackgroundFew-Shot Class-Incremental Learning from an Open-Set Perspective
Peng, C., Zhao, K., Wang, T., Li, M., & Lovell, B. C. (2022) · Lecture Notes in Computer Science · doi:10.1007/978-3-031-19806-9_22 · Cited by 110
Uncertainty-based Continual Learning with Adaptive Regularization
Ahn, H., Cha, S., Lee, D.-G., & Moon, T. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.11614 · Cited by 110 · cited 4×
“…[149],[152],spacereservationforsubsequenttasks[115],[116], [256],etc. Ontheotherhand,solutiongeneralizabilitycould beexplicitlyimprovedbyoptimizingtheflatnessmetric[81],…”
BackgroundInfluentialOn Leveraging Pretrained GANs for Generation with Limited DataPreprint
Zhao, M., Cong, Y., & Carin, L. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.11810 · Cited by 108
Flattening Sharpness for Dynamic Gradient Projection Memory Benefits Continual LearningPreprint
Deng, D., Chen, G., Hao, J., Wang, Q., & Heng, P. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.04593 · Cited by 108 · cited 9×
“When task i needs to accommodate another task j , the maximum sacrifice of its performance can be estimated by performing a second-order Taylor expansion of (cid:4) i ( θ ; D where where λ max i is the maximum eigenvalue of the Hessian matrix Note that the order of task i and j can be arbitrary, that is, (13) demonstra…”
BackgroundMethodInfluentialReplay in Deep Learning: Current Approaches and Missing Biological Elements
Hayes, T. L., Krishnan, G. P., Bazhenov, M., Siegelmann, H. T., Sejnowski, T. J., & Kanan, C. (2021) · Neural Computation · doi:10.1162/neco_a_01433 · Cited by 106 · cited 1×
“In contrast, the latest surveystypicallycollateonlyalocalaspectofcontinuallearning, withrespecttoitsbiologicalunderpinnings[1],[3],[9],[10], specializedsettingsforvisualclassification[11],[12],[13],[14], andspecificextensionsforNLP[15],[16]orRL[17].”
Note on the quadratic penalties in elastic weight consolidation
Huszár, F. (2018) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1717042115 · Cited by 105
Margin-Based Few-Shot Class-Incremental Learning with Class-Level Overfitting Mitigation
Zou, Y., Zhang, S., Li, Y., & Li, R. (2022) · neural information processing systems · doi:10.52202/068431-1977 · Cited by 105
CLiMB: A Continual Learning Benchmark for Vision-And-Language Tasks
Srinivasan, T., Chang, T.-Y., Alva, L. L. P., Chochlakis, G., Rostami, M., & Thomason, J. (2022) · neural information processing systems · doi:10.52202/068431-2135 · Cited by 104
Reparameterizing Convolutions for Incremental Multi-Task Learning Without Task Interference
Kanakis, M., Brüggemann, D., Saha, S., Georgoulis, S., Obukhov, A., & Gool, L. V. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58565-5_41 · Cited by 103
Towards Better Plasticity-Stability Trade-off in Incremental Learning: A Simple Linear Connector
Lin, G., Chu, H., & Lai, H. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.00019 · Cited by 101 · cited 4×
“…of sequentially arrived tasks within the inner loop, and exploitation of mode connectivity and flat minima in loss landscape (adapted from [86], [113]). θ ∗ A , θ ∗ B and θ ∗ A,B are desirable solutions for task A , task B and both of them, respectively. and MER [131] constrain parameter…”
BackgroundMethodInfluentialA Neural Dirichlet Process Mixture Model for Task-Free Continual Learning
Lee, S., Ha, J., Zhang, D., & Kim, G. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2001.00689 · Cited by 100 · cited 2×
“CN-DPM [80] adopts Dirichlet process mixture models to construct a growing number of neural experts, while a concurrent work [269] derives such mixture models from a probabilistic meta-learner.”
BackgroundMethodSSUL: Semantic Segmentation with Unknown Label for Exemplar-based Class-Incremental LearningPreprint
Cha, S., Kim, B., Yoo, Y., & Moon, T. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.11562 · Cited by 100 · cited 4×
“PLOP [292], RECALL [293], SSUL [294], Self-Training [295] and WILSON [296] explicitly apply the old model to generate pseudo-labels of the old classes.”
BackgroundMethodInfluentialLee, K. S., Tran, N.-T., & Cheung, N. M. (2021) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv48630.2021.00399 · Cited by 99 · cited 7×
“Similarly, resolving within-task catastrophic forgetting can facilitate reinforcement learning [299], [300] and stabilize the training of GANs [301], [302].”
BackgroundInfluentialContinual Learning with Bayesian Neural Networks for Non-Stationary Data
Kurle, R., Cseke, B., Klushyn, A., van der Smagt, P., & Günnemann, S. (2020) · International Conference on Learning Representations · Cited by 99
S-Prompts Learning with Pre-trained Transformers: An Occam's Razor for Domain Incremental LearningPreprint
Wang, Y., Huang, Z., & Hong, X. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2207.12819 · Cited by 99 · cited 1×
“…optimizing a weighted summation of the prompt pool with attention factors (CODA-Prompt [206]), using explicitly task-sharing and task-specific prompts (DualPrompt [207]) or only task-specific prompts (S-Prompts [208]), progressive expansion of task-specific prompts (Progressive Prompts [209]), etc.”
MethodLearning to Learn without Forgetting by Maximizing Transfer and Minimizing Interference
Riemer, M., Cases, I., Ajemian, R., Liu, M., Rish, I., Tu, Y., & Tesauro, G. J. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1810.11910 · Cited by 98 · cited 4×
“Forexample, MER[131]alignstheirgradientdirections,whileiTAML[184] appliesameta-updatingruletokeeptheminbalancewitheach other.”
BackgroundInfluentialLearning Latent Representations Across Multiple Data Domains Using Lifelong VAEGAN
Ye, F., & Bors, A. G. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58565-5_46 · Cited by 97
Continual pre-training mitigates forgetting in language and vision
Cossu, A., Carta, A., Passaro, L., Lomonaco, V., Tuytelaars, T., & Bacciu, D. (2024) · Neural Networks · doi:10.1016/j.neunet.2024.106492 · Cited by 97
AFEC: Active Forgetting of Negative Transfer in Continual LearningPreprint
Wang, L., Zhang, M., Jia, Z., Li, Q., Bao, C., Ma, K., Zhu, J., & Zhong, Y. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.12187 · Cited by 97 · cited 11×
“Of note, the above analyses are mainly based on finding a shared solution for all incremental tasks, which is subject to severe inter-task interference [353], [429], [430].”
BackgroundInfluentialSelf-citationBeyond Not-Forgetting: Continual Learning with Backward Knowledge Transfer
Lin, S., Yang, L., Fan, D.-L., & Zhang, J. (2022) · Neural Information Processing Systems · doi:10.48550/arxiv.2211.00789 · Cited by 97
ELLE: Efficient Lifelong Pre-training for Emerging Data
Qin, Y., Zhang, J.-J., Lin, Y.-K., et al. (2022) · Findings · doi:10.48550/arxiv.2203.06311 · Cited by 96
Reconciling meta-learning and continual learning with online mixtures of tasksPreprint
Jerfel, G., Grant, E., Griffiths, T. L., & Heller, K. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1812.06080 · Cited by 95
Episodic Memory in Lifelong Language Learning
d'Autume, C. D. M., Ruder, S., Kong, L., & Yogatama, D. (2019) · arXiv · doi:10.48550/arxiv.1906.01076 · Cited by 95 · cited 2×
“The applications of continual learning in NLP are also highly diverse, such as dialogue system [130], [131], [260], [260], [282], [298], [437], text classification [87], [172], [325], [460], sentence generation [246], [298], [499], relation learning [343], [357], [455], [465], neural machine translation [51], [128], […”
BackgroundGAN Memory with No ForgettingPreprint
Cong, Y., Zhao, M., Li, J., Wang, S., & Carin, L. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.07543 · Cited by 93 · cited 2×
“GAN-Memory [167] takes advantage of FiLM [201] and AdaFM [202] to learn task-specific parameters for each layer of a pre-trained generative model, while Authorized licensed use limited to the terms of the applicable license agreement with IEEE.”
BackgroundMethodContinual Learning of a Mixed Sequence of Similar and Dissimilar TasksPreprint
Ke, Z., Liu, B., & Huang, X. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2112.10017 · Cited by 92 · cited 1×
“…realistic consider-ationshavebeenincorporated,suchasmultiplelabels [29],noisy labels [30], blurred boundary [31], [32], hierarchical granularity [33] and sub-populations [34], mixture of task similarity [35], long-tailed distribution [36], novel class discovery [37], [38], multi-modality [39], etc.”
BackgroundMemory Efficient Continual Learning with Transformers
Ermiş, B., Zappella, G., Wistuba, M., Rawal, A., & Archambeau, C. (2022) · neural information processing systems · doi:10.52202/068431-0772 · Cited by 92 · cited 6×
“Representative strategies include prompting [402], [457], [458] or carefully fine-tuning [303], [389], [504] the backbone, saving prototypes to rectify the predictions [188], [335], [504], incorporating transfer learning techniques [43], [114], [330], etc.”
BackgroundMethodInfluentialIncorporating neuro-inspired adaptability for continual learning in artificial intelligence
Wang, L., Zhang, X., Li, Q., Zhang, M., Su, H., Zhu, J., & Zhong, Y. (2023) · Nature Machine Intelligence · doi:10.1038/s42256-023-00747-w · Cited by 91 · cited 2×
“2(e)), i.e., ∪ kt =1 θ t = ∅ , the generalization bounds are further refined by assuming K parameter spaces ( K ≥ 1 in general) to capture all tasks [69], [103].”
Self-citationUnderstanding the Role of Training Regimes in Continual Learning
Mirzadeh, S. I., Farajtabar, M., Pascanu, R., & Ghasemzadeh, H. G. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.06958 · Cited by 91 · cited 6×
“The alternatives include using an approximation of the Hessian [81], [85] or calculating δ only along the trajectory of forward and backward parameter changes [86], [87].”
BackgroundMethodInfluentialContinual Learning for Affective Robotics: Why, What and How?
Churamani, N., Kalkan, S., & Güneş, H. (2020) · IEEE RO-MAN · doi:10.1109/ro-man47096.2020.9223564 · Cited by 90
Incremental Learning in Semantic Segmentation from Image Labels
Cermelli, F., Fontanel, D., Tavera, A., Ciccone, M., & Caputo, B. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.00433 · Cited by 89 · cited 3×
“PLOP [292], RECALL [293], SSUL [294], Self-Training [295] and WILSON [296] explicitly apply the old model to generate pseudo-labels of the old classes.”
MethodClass-Incremental Novel Class Discovery
Roy, S., Liu, M., Zhong, Z., Sebe, N., & Ricci, E. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19827-4_19 · Cited by 89
Continual Learning via Local Module CompositionPreprint
Ostapenko, O., Rodríguez, P., Caccia, M., & Charlin, L. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2111.07736 · Cited by 88 · cited 1×
“MNTDP [245] and LMC [246] seek to explicitly find the optimal layout from previous sub-modules and potentially new sub-modules.”
BackgroundOnline Continual Learning Under Extreme Memory Constraints
Fini, E., Lathuilière, S., Sangineto, E., Nabi, M., & Ricci, E. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58604-1_43 · Cited by 87
Wanderlust: Online Continual Object Detection in the Real World
Wang, J., Wang, X. Y., Shangguan, Y., & Gupta, A. (2021) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.01065 · Cited by 87
CoSCL: Cooperation of Small Continual Learners is Stronger Than a Big One
Wang, L., Zhang, X., Li, Q., Zhu, J., & Zhong, Y. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19809-0_15 · Cited by 86
Self-citationGeneralisation Guarantees for Continual Learning with Orthogonal Gradient DescentPreprint
Bennani, M., Doan, T., & Sugiyama, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.11942 · Cited by 86 · cited 1×
“Sometheoreticalstudieswiththeneuraltangent kernel (NTK) [96], [97], [98], [99] suggest that an increase in task similarity may lead to more forgetting.”
BackgroundLook-ahead meta-learning for continual learning
Gupta, G. (2020) · doi:10.71781/10114 · Cited by 86
Posterior Meta-Replay for Continual LearningPreprint
Henning, C., Cervera, M. R., D'Angelo, F., von Oswald, J., Traber, R., Ehret, B., Kobayashi, S., Grewe, B. F., & Sacramento, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.01133 · Cited by 86 · cited 6×
“This is closely related to the out-of-distribution (OOD) detection, where the predictive uncertainty should be low for in-distribution data and high for OOD data [82], [160], [217].”
BackgroundInfluentialContinual Learning in Generative Adversarial Nets
Seff, A., Beatson, A., Suo, D., & Liu, H. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1705.08395 · Cited by 86
Architecture Matters in Continual LearningPreprint
Mirzadeh, S. I., Chaudhry, A., Yin, D., Nguyen, T., Pascanu, R., Görür, D., & Farajtabar, M. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2202.00275 · Cited by 86
First Session Adaptation: A Strong Replay-Free Baseline for Class-Incremental Learning
Panos, A., Kobe, Y., Reino, D. O., Aljundi, R., & Turner, R. E. (2023) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51070.2023.01725 · Cited by 85 · cited 2×
“Besides, by saving prototypes, appendinganearestclassmean(NCM)classifiertothebackbone hasprovedtobeastrongbaseline[210],[211],whichcanbe furtherenhancedbytransferlearningtechniquessuchasthe FiLMadapter[212].Asforoptimizingan Asforoptimizingan updatable backbone, F2M [84] searches for flat local minima in the pre-trainin…”
BackgroundOnline Learned Continual Compression with Adaptive Quantization ModulesPreprint
Caccia, L., Belilovsky, E., Caccia, M., & Pineau, J. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1911.08019 · Cited by 84
DualNet: Continual Learning, Fast and Slow
Pham, Q., Liu, C., & Hoi, S. (2021) · arXiv · doi:10.48550/arxiv.2110.00175 · Cited by 84
Incremental Object Detection via Meta-Learning
Joseph, K. J., Rajasegaran, J., Khan, S. H., Khan, F. S., & Balasubramanian, V. N. (2021) · Research Archive of Indian Institute of Technology Hyderabad (Indian Institute of Technology Hyderabad) · Cited by 83
Continual Learning with Foundation Models: An Empirical Study of Latent ReplayPreprint
Ostapenko, O., Lesort, T., Rodríguez, P., Arefin, M. R., Douillard, A., Rish, I., & Charlin, L. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2205.00329 · Cited by 82
Long-Tailed Class Incremental Learning
Liu, X., Hu, Y., Cao, X., Bagdanov, A. D., Li, K., & Cheng, M. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19827-4_29 · Cited by 81
Efficient Continual Learning with Modular Networks and Task-Driven PriorsPreprint
Véniat, T., Denoyer, L., & Ranzato, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2012.12631 · Cited by 81 · cited 2×
“MNTDP [416] and LMC [318] seek to explicitly find the optimal layout from previous sub-modules and”
BackgroundFunctional Regularisation for Continual Learning with Gaussian Processes
Titsias, M. K., Schwarz, J., Matthews, A., Pascanu, R., & Teh, Y. W. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1901.11356 · Cited by 81 · cited 3×
“Meanwhile, sequential Bayesian inference over function space can be seen as a form of function regularization, which generally requires storing some old training samples (called “coreset” in literature), such as FRCL [127], FROMP [128] and S-FSVI [60].”
BackgroundMethodOnline Continual Learning on a Contaminated Data Stream with Blurry Task Boundaries
Bang, J., Koh, H., Park, S., Song, H., Ha, J.-W., & Choi, J. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.00906 · Cited by 81
Continual Prompt Tuning for Dialog State Tracking
Zhu, Q., Li, B., Mi, F., Zhu, X., & Huang, M. (2022) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2022.acl-long.80 · Cited by 80
Continual Learning via Neural Pruning
Golkar, S., Kagan, M., & Cho, K. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1903.04476 · Cited by 79 · cited 2×
“PackNet [224], UCL [115], CLNP [225] and AGS-CL [116] explicitly identify the important neurons or parameters for the current task and then release the unimportant parts to the following tasks, which can be achieved by iterative pruning [224], activation value [116], [225], [226], uncertainty…”
BackgroundTriple-Memory Networks: A Brain-Inspired Method for Continual Learning
Wang, L., Lei, B., Li, Q., Su, H., Zhu, J., & Zhong, Y. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3111019 · Cited by 79 · cited 3×
“Weight regularization [25], [55], [161], [162] and experience replay [25], [163] have been shown to be effective in mitigating catastrophic forgetting of generative models.”
BackgroundSelf-citationSelf-Training for Class-Incremental Semantic Segmentation
Yu, L., Liu, X., & van de Weijer, J. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3155746 · Cited by 78 · cited 2×
“PLOP [292], RECALL [293], SSUL [294], Self-Training [295] and WILSON [296] explicitly apply the old model to generate pseudo-labels of the old classes.”
Detection and Continual Learning of Novel Face Presentation Attacks
Rostami, M. G., Spinoulas, L., Hussein, M. E., Mathai, J., & AbdAlmageed, W. (2021) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.01458 · Cited by 78 · cited 1×
“NACL [367] enables face antispoofing systems to detect potential novel attacks and remember the old ones with experience replay.”
BackgroundContinual Unsupervised Representation Learning
Rao, D., Visin, F., Rusu, A., Teh, Y. W., Pascanu, R., & Hadsell, R. T. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1910.14481 · Cited by 78 · cited 1×
“According to the provided X t and Y t in each D t , continual learning is further extended to zero-shot [23], few-shot [24], semi-supervised [25], open-world [26] and un-/self-supervised [27], [28] scenarios.”
MethodContinual Pre-Training Mitigates Forgetting in Language and VisionPreprint
Cossu, A., Tuytelaars, T., Carta, A., Passaro, L., Lomonaco, V., & Bacciu, D. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2205.09357 · Cited by 78
Prototype-Guided Continual Adaptation for Class-Incremental Unsupervised Domain Adaptation
Lin, H., Zhang, Y., Qiu, Z., Niu, S., Gan, C., Liu, Y., & Tan, M. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19827-4_21 · Cited by 78
Rehearsal revealed: The limits and merits of revisiting samples in continual learning
Verwimp, E., De Lange, M., & Tuytelaars, T. (2021) · 2021 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00925 · Cited by 75
Gradient based sample selection for online continual learning
Aljundi, R., Lin, M., Goujaud, B., & Bengio, Y. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1903.08671 · Cited by 75 · cited 3×
“Correspondingly, GCL consists of two interconnected settings: Task-Free Continual Learning (TFCL) [20], where the task identities are not accessible in either training or testing; and Online Continual Learning (OCL) [21], where the training samples are observed in an one-pass data stream.”
BackgroundNatural continual learning: success is a journey, not (just) a destination
Kao, T.-C., Jensen, K. T., van de Ven, G. M., Bernacchia, A., & Hennequin, G. (2021) · Lirias · doi:10.48550/arxiv.2106.08085 · Cited by 75 · cited 4×
“1: k − 1 ) as a multivariate Gaussian with local gradient information [49], [50], [51], [52].”
BackgroundMethodInfluentialContinual Learning for Text Classification with Information Disentanglement Based Regularization
Huang, Y., Zhang, Y., Chen, J., Wang, X., & Yang, D. (2021) · Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies · doi:10.18653/v1/2021.naacl-main.218 · Cited by 74
Incremental Few-Shot Instance Segmentation
Ganea, D. A., Boom, B., & Poppe, R. (2021) · 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr46437.2021.00124 · Cited by 74
Kukleva, A., Kuehne, H., & Schiele, B. (2021) · 2021 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00889 · Cited by 73 · cited 1×
“The extremely limited training samples exacerbate the overfitting of previously-learned representations to subsequent tasks, which can be alleviated by recent work such as preserving the topology of representations (TOPIC [404]), constructing an exemplar relation graph for knowledge distillation (ERL [97]), selectivel…”
BackgroundCPR: Classifier-Projection Regularization for Continual LearningPreprint
Cha, S., Hsu, H., Hwang, T., Calmon, F. P., & Moon, T. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.07326 · Cited by 72 · cited 4×
“A distinctive feature of CRL is its high diversity of applications and benchmarks, including continuous control [203], [279], [296], [449], maze navigation [274], [370], video games (Atari [10], [59], [195], [220], [370], [429], StarCraft [372], Minecraft [405], Hanabi [312]), etc.”
BackgroundInfluentialOn the Effectiveness of Lipschitz-Driven Rehearsal in Continual Learning
Bonicelli, L., Boschini, M., Porrello, A., Spampinato, C., & Calderara, S. (2022) · neural information processing systems · doi:10.52202/068431-2311 · Cited by 72
ScaIL: Classifier Weights Scaling for Class Incremental Learning
Belouadah, E., & Popescu, A. (2020) · 2020 IEEE Winter Conference on Applications of Computer Vision (WACV) · doi:10.1109/wacv45572.2020.9093562 · Cited by 72
Online Continual Learning on Class Incremental Blurry Task Configuration with Anytime InferencePreprint
Koh, H., Kim, D., Ha, J.-W., & Choi, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.10031 · Cited by 71 · cited 6×
“entropy (ER-ACE [47]) or constrained gradient directions (GEM [277], A-GEM [65]) of the old and new training samples, repeated rehearsal with data augmentation (RAR [498]), properly adjusting the learning rate (PoLRS [50], CLIB [224]), training from scratch with the memory buffer”
BackgroundInfluentialECONET: Effective Continual Pretraining of Language Models for Event Temporal Reasoning
Han, R., Ren, X., & Peng, N. (2021) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2021.emnlp-main.436 · Cited by 71
Novel Class Discovery Without Forgetting
Joseph, K. J., Paul, S., Aggarwal, G., Biswas, S., Rai, P., Han, K., & Balasubramanian, V. N. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-20053-3_33 · Cited by 70
Energy-based Latent Aligner for Incremental Learning
Joseph, K. J., Khan, S. H., Khan, F. S., Anwer, R. M., & Balasubramanian, V. N. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.00730 · Cited by 70 · cited 1×
“ELI [151] learns an energy manifold with the old and new models to realign the representation shifts for optimizing incremental tasks.”
MethodShao, C., & Feng, Y. (2022) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2022.acl-long.143 · Cited by 70
Learn to Grow: A Continual Structure Learning Framework for Overcoming Catastrophic Forgetting
Li, X., Zhou, Y., Wu, T., Socher, R., & Xiong, C. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1904.00310 · Cited by 69 · cited 1×
“reinforcement learning (RCL [464], BNS [344]), architecture search (LtG [251], BNS [344]), variational Bayes (BSA [230]), etc.”
BackgroundRandom Path Selection for Incremental LearningPreprint
Rajasegaran, J., Hayat, M., Khan, S. H., Khan, F. S., & Shao, L. (2019) · arXiv (Cornell University) · Cited by 69
Policy Consolidation for Continual Reinforcement LearningPreprint
Kaplanis, C., Shanahan, M., & Clopath, C. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1902.00255 · Cited by 69 · cited 3×
“A distinctive feature of CRL is its high diversity of applications and benchmarks, including continuous control [203], [279], [296], [449], maze navigation [274], [370], video games (Atari [10], [59], [195], [220], [370], [429], StarCraft [372], Minecraft [405], Hanabi [312]), etc.”
BackgroundContinual Learning on Noisy Data Streams via Self-Purified Replay
Kim, C. D., Jeong, J., Moon, S., & Kim, G. (2021) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00058 · Cited by 68 · cited 2×
“To cope with noisy labeled incremental data, SPR [216] obtains self-supervised representations to purify a memory buffer and then performs supervised finetuning.”
BackgroundMethodBoschini, M., Bonicelli, L., Porrello, A., Bellitto, G., Pennisi, M., Palazzo, S., Spampinato, C., & Calderara, S. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-20050-2_40 · Cited by 68
On the Effectiveness of Lipschitz-Driven Rehearsal in Continual Learning
Bonicelli, L., Boschini, M., Porrello, A., Spampinato, C., & Calderara, S. (2022) · Neural Information Processing Systems · doi:10.48550/arxiv.2210.06443 · Cited by 68
Continual Few-shot Relation Learning via Embedding Space Regularization and Data Augmentation
Qin, C., & Joty, S. (2022) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2022.acl-long.198 · Cited by 68
Continuous Scene Representations for Embodied AI
Gadre, S. Y., Ehsani, K., Song, S., & Mottaghi, R. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.01443 · Cited by 68
The Challenges of Continuous Self-Supervised Learning
Purushwalkam, S., Morgado, P. M., & Gupta, A. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19809-0_40 · Cited by 67
Overcoming Catastrophic Forgetting by Neuron-Level Plasticity Control
Paik, I., Oh, S., Kwak, T.-Y., & Kim, I. (2020) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v34i04.5981 · Cited by 66
Bring Evanescent Representations to Life in Lifelong Class Incremental Learning
Toldo, M., & Özay, M. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.01623 · Cited by 65 · cited 2×
“Other methods exploit the class-wise statistics of feature representations to obtain a balanced classifier, such as by imposing the representations to be transferable and invariant [173], [176], or compensating explicitly the representation shifts [174], [249].”
BackgroundMethodSame State, Different Task: Continual Reinforcement Learning without Interference
Kessler, S., Parker-Holder, J., Ball, P., Zohren, S., & Roberts, S. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i7.20674 · Cited by 65
Adaptive Orthogonal Projection for Batch and Online Continual Learning
Guo, Y., Hu, W., Zhao, D., & Liu, B. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i6.20634 · Cited by 65
Layerwise Optimization by Gradient Decomposition for Continual Learning
Tang, S., Chen, D., Zhu, J., Yu, S., & Ouyang, W. (2021) · 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr46437.2021.00951 · Cited by 64 · cited 5×
“LOGD [401] decomposes the gradient of each task into task-sharing and task-specific components to leverage inter-task information.”
BackgroundMethodInfluentialSelf-citationSynthesized Feature based Few-Shot Class-Incremental Learning on a Mixture of Subspaces
Cheraghian, A., Rahman, S., Ramasinghe, S., Fang, P., Simon, C., Petersson, L., & Harandi, M. (2021) · 2021 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00854 · Cited by 64
Origins of Cell-Type-Specific Olfactory Processing in the Drosophila Mushroom Body Circuit
Inada, K., Tsuchimoto, Y., & Kazama, H. (2017) · Neuron · doi:10.1016/j.neuron.2017.06.039 · Cited by 64
Continual Normalization: Rethinking Batch Normalization for Online Continual LearningPreprint
Pham, Q., Liu, C., & Hoi, S. C. H. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2203.16102 · Cited by 64
Mimicking the Oracle: An Initial Phase Decorrelation Approach for Class Incremental Learning
Shi, Y., Zhou, K., Liang, J., Jiang, Z., Feng, J., Torr, P., Bai, S., & Tan, V. Y. F. (2022) · 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52688.2022.01622 · Cited by 63 · cited 5×
“CwD [391] regularizes the initial-phase representations to be uniformly scattered, which can empirically mimic the representations of joint training.”
BackgroundMethodInfluentialInformation-theoretic Online Memory Selection for Continual LearningPreprint
Sun, S., Calandriello, D., Hu, H., Li, A., & Titsias, M. K. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2204.04763 · Cited by 63 · cited 1×
“More advanced strategies prioritize the replay of those training samples that are informative [273], diversified in parameter gradients [21], balanced in class labels [159], [274], and beneficial to latent decision boundaries [134].”
BackgroundGeneral Incremental Learning with Domain-aware Categorical Representations
Xie, J., Yan, S., & He, X. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.01395 · Cited by 61 · cited 1×
“Some recent work has focused on combinatorial scenarios [18], [40], [41], [42], [43] in order to simulate real-world complexity.”
BackgroundThe Challenges of Continuous Self-Supervised Learning
Purushwalkam, S., Morgado, P., & Gupta, A. (2022) · European Conference on Computer Vision · doi:10.48550/arxiv.2203.12710 · Cited by 61
Zhang, X., Li, Q., Wang, L., Liu, Z., & Zhong, Y. (2018) · Neuron · doi:10.1016/j.neuron.2018.02.025 · Cited by 61
Self-citationBalancing Stability and Plasticity Through Advanced Null Space in Continual Learning
Kong, Y., Liu, L., Wang, Z., & Tao, D. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19809-0_13 · Cited by 60
Learning with Recoverable Forgetting
Ye, J., Fu, Y., Song, J., Yang, X., Liu, S., Jin, X., Song, M., & Wang, X. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-20083-0_6 · Cited by 60
Exemplar-Supported Generative Reproduction for Class Incremental Learning.
(2018) · British Machine Vision Conference (BMVC) · Cited by 60
Beaulieu, S., Frati, L., Miconi, T., Lehman, J., Stanley, K. O., Clune, J., & Cheney, N. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.09571 · Cited by 60 · cited 2×
“AIM[183]learnsamixtureofexpertstomakepredic-tionswiththerepresentationsofOML[181]orANML[182], furthersparsifyingtherepresentationsatthearchitecturallevel.”
BackgroundMeta-Consolidation for Continual LearningPreprint
Joseph, K. J., & Balasubramanian, V. N. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.00352 · Cited by 60 · cited 3×
“To address this issue, the i th task’s distribution p ( D t | i, θ ) could be recovered by replay [74], [78], [79], [80] p ( where the marginal task distribution p ( i ) ∝ N i in general.”
BackgroundMethodGreco, C., Plank, B., Fernández, R., & Bernardi, R. (2019) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/p19-1350 · Cited by 60
ALIFE: Adaptive Logit Regularizer and Feature Replay for Incremental Semantic Segmentation
Oh, Y., Baek, D., & Ham, B. (2022) · neural information processing systems · doi:10.52202/068431-1055 · Cited by 59
Continual SLAM: Beyond Lifelong Simultaneous Localization and Mapping Through Continual Learning
Vödisch, N., Cattaneo, D., Burgard, W., & Valada, A. (2023) · Springer proceedings in advanced robotics · doi:10.1007/978-3-031-25555-7_3 · Cited by 58
Generalized Variational Continual LearningPreprint
Loo, N., Swaroop, S., & Turner, R. E. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.12328 · Cited by 58 · cited 4×
“lel branches (ACL [107], ReduNet [459], EPIE-Net [104]), adaptive layers (GVCL [276], DyTox [101]), masks or mask”
BackgroundInfluentialDong, T., He, J., Wang, S., Wang, L., Cheng, Y., & Zhong, Y. (2016) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1602152113 · Cited by 58
Self-citationContinual Sequence Generation with Adaptive Compositional Modules
Zhang, Y., Wang, X., & Yang, D. (2022) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2022.acl-long.255 · Cited by 58
Continual Learning With Extended Kronecker-Factored Approximate Curvature
Lee, J., Hong, H. G., Joo, D., & Kim, J. (2020) · 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr42600.2020.00902 · Cited by 57
Mitigating Forgetting in Online Continual Learning via Instance-Aware Parameterization
Chen, H., Cheng, A., Juan, D.-C., Wei, W., & Sun, M. (2020) · neural information processing systems · Cited by 57
Towards Continual Learning for Multilingual Machine Translation via Vocabulary Substitution
Garcia, X., Constant, N., Parikh, A. P., & Fırat, O. (2021) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2021.naacl-main.93 · Cited by 57
Continual Learning by Asymmetric Loss Approximation With Single-Side Overestimation
Park, D.-M., Hong, S., Han, B., & Lee, K. M. (2019) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00343 · Cited by 56 · cited 2×
“Observing the asymmetric effect of parameter changes on old tasks, ALASSO [110] designs an asymmetric quadratic penalty with one of its sides overestimated.”
BackgroundRemembering for the right reasons: Explanations reduce catastrophic forgetting
Ebrahimi, S., Petryk, S., Gokul, A., Gan, W., Gonzalez, J. E., Rohrbach, M., & Darrell, T. (2021) · Applied AI Letters · doi:10.1002/ail2.44 · Cited by 56
Improving and Understanding Variational Continual LearningPreprint
Swaroop, S., Nguyen, C. V., Bui, T. D., & Turner, R. E. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.02099 · Cited by 56
Generalized Class Incremental Learning
Mi, F., Kong, L., Lin, T., Yu, K., & Faltings, B. V. (2020) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw50498.2020.00128 · Cited by 55 · cited 1×
“Some recent work has focused on combinatorial scenarios [18], [40], [41], [42], [43] in order to simulate real-world complexity.”
BackgroundIncremental Learning via Rate Reduction
Wu, Z., Baek, C., You, C., & Ma, Y. (2021) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr46437.2021.00118 · Cited by 55 · cited 1×
“For a regular network, the task-specific components could be parallel branches [233], [234], adaptive layers [58], [235], masks of intermediate features [186], [236], [237].”
BackgroundContinual semi-supervised learning through contrastive interpolation consistency
Boschini, M., Buzzega, P., Bonicelli, L., Porrello, A., & Calderara, S. (2022) · Pattern Recognition Letters · doi:10.1016/j.patrec.2022.08.006 · Cited by 55 · cited 1×
“Subsequent work includes training an adversarial autoencoder to reconstruct images [265], imposing predictive consistency among augmented and interpolated data [266], and leveraging the nearest-neighbor classifier to distill class-instance relationships [267].”
MethodProbing Representation Forgetting in Supervised and Unsupervised Continual Learning
Davari, M., Asadi, N., Mudur, S., Aljundi, R., & Belilovsky, E. (2022) · 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52688.2022.01621 · Cited by 55 · cited 3×
“In particular, the benefits for downstream continual learning tend to be more apparent when pre-training with larger data size [316], [352], larger model size [352] and contrastive loss [83], [125].”
BackgroundLearn Continually, Generalize Rapidly: Lifelong Knowledge Accumulation for Few-shot Learning
Jin, X., Lin, B. Y., Rostami, M. G., & Ren, X. (2021) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2021.findings-emnlp.62 · Cited by 55
Shi, G., Wu, Y., Liu, J., Wan, S., Wang, W., & Lü, T. (2022) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3503161.3548218 · Cited by 55
Image De-raining via Continual Learning
Zhou, M., Xiao, J., Chang, Y., Fu, X., Liu, A., Pan, J., & Zha, Z.-J. (2021) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr46437.2021.00487 · Cited by 54 · cited 1×
“Asastrategyforadaptingtovariableinputs,continuallearn-ingcanassistarobustmodeltoeliminateorresistexternal disturbances[30],[313],[314].”
Contextual Transformation Networks for Online Continual Learning
Pham, Q., Liu, C., Sahoo, D., & Hoi, S. C. H. (2021) · Singapore Management University Institutional Knowledge (InK) (Singapore Management University) · Cited by 54
Memory Replay GANs: learning to generate images from new categories without forgetting
Wu, C., Herranz, L., Liu, X., Wang, Y., van de Weijer, J., & Raducanu, B. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1809.02058 · Cited by 54
Continual Learning in Low-rank Orthogonal Subspaces
Chaudhry, A., Khan, N., Dokania, P. K., & Torr, P. H. S. (2020) · arXiv · doi:10.48550/arxiv.2010.11635 · Cited by 54 · cited 2×
“Orthog-Subspace [64] performs continual learning with orthogonal low-rank vector subspaces and keeps the”
BackgroundHyper-LifelongGAN: Scalable Lifelong Learning for Image Conditioned Generation
Zhai, M., Chen, L., & Mori, G. (2021) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr46437.2021.00228 · Cited by 54
Continual Deep Learning by Functional Regularisation of Memorable Past
Pan, P., Swaroop, S., Immer, A., Eschenhagen, R., Turner, R. E., & Khan, M. E. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.14070 · Cited by 53 · cited 1×
“Meanwhile, sequential Bayesian inference over function space can be seen as a form of function regularization, which generally requires storing some old training samples (called “coreset” in literature), such as FRCL [127], FROMP [128] and S-FSVI [60].”
BackgroundEEC: Learning to Encode and Regenerate Images for Continual LearningPreprint
Ayub, A., & Wagner, A. R. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2101.04904 · Cited by 53 · cited 4×
“In contrast, autoencoder-based strategies, such as FearNet [210], SRM [361], CLEER [366], EEC [26], GMR [334] and Flashcards [138], can explicitly control the labels of the generated data, although relatively blurred.”
BackgroundMethodInfluentialCalibrating CNNs for Lifelong Learning
Singh, P. P., Verma, V. K., Mazumder, P., Carin, L., & Rai, P. (2020) · neural information processing systems · Cited by 53
Continual Learning through Retrieval and Imagination
Wang, Z., Liu, L., Duan, Y., & Tao, D. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i8.20837 · Cited by 51 · cited 3×
“…tasksolutions[52],[104],[113],[154],[196],balancedexploita-tionofoldandnewtrainingsamples[46],[129],[147],[148], [149],[152],spacereservationforsubsequenttasks[115],[116], [256],etc. Ontheotherhand,solutiongeneralizabilitycould…”
BackgroundHelpful or Harmful: Inter-task Association in Continual Learning
Jin, H., & Kim, E. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-20083-0_31 · Cited by 51 · cited 2×
“Also, the function of task-identity prediction can be explicitly learned from incremental tasks, using other continual learning strategies to mitigate catastrophic forgetting [1], [79], [107], [160], [189], [221], [294].”
BackgroundContinual Barlow Twins: Continual Self-Supervised Learning for Remote Sensing Semantic Segmentation
Marsocci, V., & Scardapane, S. (2023) · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · doi:10.1109/jstars.2023.3280029 · Cited by 51 · cited 2×
“The third scenario is to learn representations from incremental unlabeled data [28], [215], [216], which becomes an increasingly important topic for updating pre-trained knowledge in foundation models.”
Wide Neural Networks Forget Less CatastrophicallyPreprint
Mirzadeh, S. I., Chaudhry, A., Yin, D., Hu, H., Pascanu, R., Görür, D., & Farajtabar, M. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.11526 · Cited by 50 · cited 1×
“The third comes down to obtaining well-distributed representations, which tend to be more robust to distribution differences in function space, such as by using large-scale pre-training [87], [89], [90] and self-supervised learning [28], [91], [92], [93].”
BackgroundModel Zoo: A Growing "Brain" That Learns ContinuallyPreprint
Ramesh, R., & Chaudhari, P. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.03027 · Cited by 50 · cited 4×
“The generalization error for each task can improve with synergistic tasks but deteriorate with competing tasks [68], where learning all tasks equally in a shared solution tends to compromise each task in performance [68], [69].”
BackgroundInfluentialConTinTin: Continual Learning from Task Instructions
Yin, W., Li, J., & Xiong, C. (2022) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2022.acl-long.218 · Cited by 50
Class-Incremental Learning with Cross-Space Clustering and Controlled Transfer
Ashok, A., Joseph, K. J., & Balasubramanian, V. N. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19812-0_7 · Cited by 49
Efficient Meta Lifelong-Learning with Limited Memory
Wang, Z., Mehta, S. V., Poczos, B., & Carbonell, J. (2020) · Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2020.emnlp-main.39 · Cited by 49
Scalable Recollections for Continual Lifelong Learning
Riemer, M., Klinger, T., Bouneffouf, D., & Franceschini, M. (2019) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v33i01.33011352 · Cited by 48 · cited 1×
“In contrast, autoencoder-based strategies, such as FearNet [165], SRM [169] and EEC [168], can explicitly control the labels of the generated data, although relatively blurred.”
MethodContributions by metaplasticity to solving the Catastrophic Forgetting Problem
Jedlička, P., Tomko, M., Robins, A. V., & Abraham, W. C. (2022) · Trends in Neurosciences · doi:10.1016/j.tins.2022.06.002 · Cited by 48 · cited 1×
“In contrast, the latest surveystypicallycollateonlyalocalaspectofcontinuallearning, withrespecttoitsbiologicalunderpinnings[1],[3],[9],[10], specializedsettingsforvisualclassification[11],[12],[13],[14], andspecificextensionsforNLP[15],[16]orRL[17].”
NISPA: Neuro-Inspired Stability-Plasticity Adaptation for Continual Learning in Sparse NetworksPreprint
Gürbüz, M. B., & Dovrolis, C. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2206.09117 · Cited by 48 · cited 1×
“PackNet [286], UCL [10], CLNP [136], AGS-CL [195] and NISPA [147] explicitly identify the important neurons or parameters for the current task and then release the unimportant parts to the following tasks, which can be achieved by iterative pruning [286], activation value [136], [147], [195], uncertainty estimation [1…”
BackgroundThe Ideal Continual Learner: An Agent That Never ForgetsPreprint
Peng, L., Giampouras, P., & Vidal, R. (2023) · International Conference on Machine Learning · doi:10.48550/arxiv.2305.00316 · Cited by 48 · cited 1×
“For example, with probability at least 1 − δ for any δ ∈ (0 , 1) , an ideal continual learner [102] under the assumption that all tasks share a global minimizer with uniform convergence (i.e., λ max i = λ for ∀ t = 1 , . . . , k in (13)) has the generalization bound where c ∗ t = (cid:4) t ( θ ∗ t…”
Task-Free Continual Learning Via Online Discrepancy Distance Learning
Ye, F., & Bors, A. G. (2022) · neural information processing systems · doi:10.52202/068431-1720 · Cited by 48
Learning with Selective Forgetting
Shibata, T., Irie, G., Ikami, D., & Mitsuzumi, Y. (2021) · Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2021/137 · Cited by 47 · cited 1×
“MnemonicCode[311]embedsa class-specificcodewhenlearningeachclass,enablingtoselec-tivelyforgetthemthroughdiscardingthecorrespondingcodes.”
Learning to Imagine: Diversify Memory for Incremental Learning using Unlabeled Data
Tang, Y.-M., Peng, Y.-X., & Zheng, W. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.00933 · Cited by 46 · cited 1×
“Besides, other work considers a slightly different setting, where an additional unlabeled dataset is used for knowledge distillation (GD [239]) and data augmentation (L2I [402]), in order to facilitate supervised continual learning.”
BackgroundA Simple Baseline that Questions the Use of Pretrained-Models in Continual LearningPreprint
Janson, P., Zhang, W., Aljundi, R., & Elhoseiny, M. (2022) · King Abdullah University of Science and Technology Repository (King Abdullah University of Science and Technology) · doi:10.48550/arxiv.2210.04428 · Cited by 46 · cited 2×
“Representative strategies include prompting [402], [457], [458] or carefully fine-tuning [303], [389], [504] the backbone, saving prototypes to rectify the predictions [188], [335], [504], incorporating transfer learning techniques [43], [114], [330], etc.”
BackgroundMethodRectification-based Knowledge Retention for Continual Learning
Singh, P., Mazumder, P., Rai, P., & Namboodiri, V. P. (2021) · 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr46437.2021.01503 · Cited by 45 · cited 1×
“According to the provided Xt and Yt in each Dt, continual learning can be further categorized into zero-shot [132], [388], few-shot [404], semi-supervised [428], open-world (i.”
BackgroundS3C: Self-Supervised Stochastic Classifiers for Few-Shot Class-Incremental Learning
Kalla, J., & Biswas, S. (2022) · Lecture Notes in Computer Science · doi:10.1007/978-3-031-19806-9_25 · Cited by 45 · cited 2×
“feature space with virtual prototypes for future classes (Fact [504]), using angular penalty loss with data augmentation (ALICE [329]), using self-supervised learning (S3C [196]),”
BackgroundMethodTask-Free Continual Learning via Online Discrepancy Distance Learning
Ye, F., & Bors, A. (2022) · Neural Information Processing Systems · doi:10.48550/arxiv.2210.06579 · Cited by 44
RATT: Recurrent Attention to Transient Tasks for Continual Image CaptioningPreprint
Chiaro, R. D., Twardowski, B., Bagdanov, A. D., & van de Weijer, J. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2007.06271 · Cited by 44
TRGP: Trust Region Gradient Projection for Continual Learning
Lin, S., Yang, L., Fan, D., & Zhang, J. (2022) · arXiv · doi:10.48550/arxiv.2202.02931 · Cited by 43 · cited 1×
“TRGP [178] defines the “trust region” through the norm of gradient projection onto the subspace of previous inputs, so as to selectively reuse the frozen weights of old tasks.”
BackgroundOptimizing Reusable Knowledge for Continual Learning via MetalearningPreprint
Hurtado, J., Raymond-Sáez, A., & Soto, Á. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.05390 · Cited by 43 · cited 2×
“MARK [186] maintains a set of shared weights that are incrementally updated with meta-learning and selectively masked to solve specific tasks.”
BackgroundImproving Task-free Continual Learning by Distributionally Robust Memory EvolutionPreprint
Wang, Z., Shen, L., Fang, L., Suo, Q., Duan, T., & Gao, M. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2207.07256 · Cited by 42 · cited 2×
“Meanwhile, the memory buffer can be dynamically managed, such as by removing less important training samples (CLIB [224]), editing the old training samples (GMED [191]) or their distributions (DRO [443]) to be more likely forgotten, and retrieving the old training samples that are susceptible to interference (MIR [14]…”
BackgroundDeltaGrad: Rapid retraining of machine learning models
Wu, Y., Dobriban, E., & Davidson, S. B. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.14755 · Cited by 41 · cited 1×
“…are closely related to continuallearning,suchaslearningseparatemodelswithsubsets oftrainingsamples[307],utilizinghistoricalparametersand gradients[308],removingoldknowledgefromparameterswith Fisherinformationmatrix[309],addingadaptiveparameterson apre-trainedmodel[310],etc.…”
DyGRAIN: An Incremental Learning Framework for Dynamic Graphs
Kim, S., Yun, S., & Kang, J. (2022) · International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2022/438 · Cited by 41
Generative Negative Text Replay for Continual Vision-Language Pretraining
Yan, S., Hong, L., Xu, H., Han, J., Tuytelaars, T., Li, Z., & He, X. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-20059-5_2 · Cited by 40
How Well Does Self-Supervised Pre-Training Perform with Streaming Data?Preprint
Hu, D., Yan, S., Lu, Q., Hong, L., Hu, H. Z., Zhang, Y., Li, Z., Wang, X., & Feng, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2104.12081 · Cited by 40 · cited 8×
“In particular,sinceself-supervisedandpre-trainedrepresentations arenaturallymorerobusttocatastrophicforgetting[28],[87], [89],[93],creating,improvingandexploitingsuchrepresenta-tionaladvantageshasbecomeapromisingdirection.”
BackgroundMethodResultInfluentialContinual Learning with Recursive Gradient OptimizationPreprint
Liu, H., & Liu, H. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2201.12522 · Cited by 40 · cited 1×
“Under the upper bound of the quadratic penalty in Bayesian weight regularization, RGO [180] modifies gradient directions with a recursive optimization procedure to obtain the optimal solution.”
MethodContinual Learning for Neural Machine Translation
Cao, Y., Wei, H., Chen, B., & Wan, X. (2021) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2021.naacl-main.310 · Cited by 40
Recent Advances of Continual Learning in Computer Vision: An Overview
(2021) · doi:10.48550/arxiv.2109.11369 · Cited by 39 · cited 1×
“In contrast, the latest surveystypicallycollateonlyalocalaspectofcontinuallearning, withrespecttoitsbiologicalunderpinnings[1],[3],[9],[10], specializedsettingsforvisualclassification[11],[12],[13],[14], andspecificextensionsforNLP[15],[16]orRL[17].”
Doodle It Yourself: Class Incremental Learning by Drawing a Few Sketches
Bhunia, A. K., Gajjala, V. R., Koley, S., Kundu, R., Sain, A., Xiang, T., & Song, Y.-Z. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.00233 · Cited by 39 · cited 1×
“Besides, auxiliary information such as semantic word vectors [262], [263] and sketch exemplars [264] can be incorporated to enrich the limited training samples.”
BackgroundPiggyback GAN: Efficient Lifelong Learning for Image Conditioned Generation
Zhai, M., Chen, L., He, J., Nawhal, M., Tung, F., & Mori, G. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58589-1_24 · Cited by 39
An Empirical Investigation of the Role of Pre-training in Lifelong Learning
Mehta, S. V., Patil, D., Chandar, S., & Strubell, E. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2112.09153 · Cited by 38 · cited 9×
“In addition to an earlier work on obtaining sparse representations through meta-training [185], recent works have attempted to incorporate the advantages of selfsupervised learning (SSL) [125], [281], [335] and large-scale pre-training [295], [380], [456] to improve the representations in initialization and in continu…”
BackgroundMethodInfluentialLearnability and Algorithm for Continual LearningPreprint
Kim, G., Xiao, C., Konishi, T., & Liu, B. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2306.12646 · Cited by 38 · cited 2×
“This is closely related to the out-of-distribution (OOD) detection, where the predictive uncertainty should be low for in-distribution data and high for OOD data [74], [75], [76], [77].”
Few-Shot and Continual Learning with Attentive Independent Mechanisms
Lee, E., Huang, C., & Lee, C. (2021) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00932 · Cited by 37 · cited 1×
“AIM[183]learnsamixtureofexpertstomakepredic-tionswiththerepresentationsofOML[181]orANML[182], furthersparsifyingtherepresentationsatthearchitecturallevel.”
A soft nearest-neighbor framework for continual semi-supervised learning
Kang, Z., Fini, E., Nabi, M., Ricci, E., & Alahari, K. (2023) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51070.2023.01090 · Cited by 37 · cited 1×
“Targeting a similar setting, subsequent work includes training an adversarial autoencoder to reconstruct images [242], imposing predictive consistency among augmented and interpolated data [44], and leveraging the nearest-neighbor classifier to distill class-instance relationships [204].”
Continual Learning with Adaptive Weights (CLAW)
Adel, T., Zhao, H., & Turner, R. E. (2019) · ENLIGHTEN (Jurnal Bimbingan dan Konseling Islam) · doi:10.48550/arxiv.1911.09514 · Cited by 37 · cited 2×
“The second is online variational inference (VI) [55], [56], [57], [58], [59], [60].”
BackgroundDo not Forget to Attend to Uncertainty while Mitigating Catastrophic Forgetting
Kurmi, V. K., Patro, B. N., Subramanian, V. K., & Namboodiri, V. P. (2021) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv48630.2021.00078 · Cited by 36
Stream-51: Streaming Classification and Novelty Detection from Videos
Roady, R., Hayes, T. L., Vaidya, H., & Kanan, C. (2020) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw50498.2020.00122 · Cited by 36
Static-Dynamic Co-teaching for Class-Incremental 3D Object Detection
Zhao, N., & Lee, G. H. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i3.20254 · Cited by 36
Continual Training of Language Models for Few-Shot Learning
Ke, Z., Lin, H., Shao, Y., Xu, H., Shu, L., & Liu, B. (2022) · Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2022.emnlp-main.695 · Cited by 35
Gong, Z., Zhou, K., Zhao, X., Sha, J. J., Wang, S., & Wen, J.-R. (2022) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2022.acl-long.408 · Cited by 34 · cited 3×
“Theauxiliaryinformationwithlowstoragecost,suchasclass statistics[79],[140]andattentionmaps[141],[142],canbe further incorporated to maintain old knowledge.”
BackgroundContinual Learning of Natural Language Processing Tasks: A Survey
Ke, Z., & Liu, B. (2022) · arXiv · doi:10.48550/arxiv.2211.12701 · Cited by 34 · cited 1×
“In contrast, the latest surveystypicallycollateonlyalocalaspectofcontinuallearning, withrespecttoitsbiologicalunderpinnings[1],[3],[9],[10], specializedsettingsforvisualclassification[11],[12],[13],[14], andspecificextensionsforNLP[15],[16]orRL[17].”
Lifelong and Continual Learning Dialogue Systems: Learning during Conversation
Liu, B., & Mazumder, S. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i17.17768 · Cited by 34
IIRC: Incremental Implicitly-Refined Classification
Abdelsalam, M. F. A., Faramarzi, M., Sodhani, S., & Chandar, S. (2021) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr46437.2021.01089 · Cited by 33
Anti-retroactive Interference for Lifelong Learning
Wang, R., Bao, Y., Zhang, B., Liu, J., Zhu, W., & Guo, G. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-20053-3_10 · Cited by 33
Optimal Continual Learning has Perfect Memory and is NP-hardPreprint
Knoblauch, J., Husain, H., & Diethe, T. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.05188 · Cited by 33 · cited 4×
“Such a classical continual learning problem has proven to be NP-hard in general [62], because the feasible parameter space tends to be narrow and irregular as more tasks are introduced, thus difficult to identify (see Fig.”
BackgroundInfluentialProgressive Prompts: Continual Learning for Language Models
Razdaibiedina, A., Mao, Y., Hou, R., Khabsa, M., Lewis, M., & Almahairi, A. (2023) · arXiv · doi:10.48550/arxiv.2301.12314 · Cited by 33 · cited 1×
“…optimizing a weighted summation of the prompt pool with attention factors (CODA-Prompt [206]), using explicitly task-sharing and task-specific prompts (DualPrompt [207]) or only task-specific prompts (S-Prompts [208]), progressive expansion of task-specific prompts (Progressive Prompts [209]), etc.”
MethodA Progressive Model to Enable Continual Learning for Semantic Slot Filling
Shen, Y., Zeng, X., & Jin, H. (2019) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/d19-1126 · Cited by 33
Incremental Prompting: Episodic Memory Prompt for Lifelong Event DetectionPreprint
Liu, M., Chang, S., & Huang, L. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2204.07275 · Cited by 32
Knowledge Capture and Replay for Continual Learning
Gopalakrishnan, S., Singh, P. R., Fayek, H. M., Ramasamy, S., & Ambikapathi, A. (2022) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv51458.2022.00041 · Cited by 32
Class Gradient Projection For Continual Learning
Chen, C., Zhang, J., Song, J., & Gao, L. (2022) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3503161.3548054 · Cited by 31
Online Task-free Continual Learning with Dynamic Sparse Distributed Memory
Pourcel, J., Vu, N., & French, R. M. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19806-9_42 · Cited by 31
Split-and-Bridge: Adaptable Class Incremental Learning within a Single Neural Network
Kim, J.-Y., & Choi, D.-W. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i9.16991 · Cited by 31
Continual Learning Beyond a Single ModelPreprint
Doan, T., Mirzadeh, S. I., & Farajtabar, M. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2202.09826 · Cited by 30
Self-Supervised Training Enhances Online Continual LearningPreprint
Gallardo, J., Hayes, T. L., & Kanan, C. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.14010 · Cited by 28 · cited 7×
“In addition to an earlier work on obtaining sparse representations through meta-training [185], recent works have attempted to incorporate the advantages of selfsupervised learning (SSL) [125], [281], [335] and large-scale pre-training [295], [380], [456] to improve the representations in initialization and in continu…”
BackgroundInfluentialKernel Continual LearningPreprint
Derakhshani, M. M., Zhen, X., Shao, L., & Snoek, C. G. M. (2021) · UvA-DARE (University of Amsterdam) · doi:10.48550/arxiv.2107.05757 · Cited by 28
Zhang, X., Zhao, T.-F., Chen, J., Shen, Y., & Li, X. (2022) · Nature Communications · doi:10.1038/s41467-022-29994-y · Cited by 28
Self-citationLee, J., Joo, D., Hong, H. G., & Kim, J. (2020) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v34i04.5884 · Cited by 27
Continual Learning using a Bayesian Nonparametric Dictionary of Weight FactorsPreprint
Mehta, N., Liang, K. J., Verma, V. K., & Carin, L. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.10098 · Cited by 27
Semi-Supervised Class Incremental Learning
Lechat, A., Herbin, S., & Jurie, F. (2021) · Proceedings - International Conference on Pattern Recognition/Proceedings/International Conference on Pattern Recognition · doi:10.1109/icpr48806.2021.9413225 · Cited by 26 · cited 1×
“Subsequent work includes training an adversarial autoencoder to reconstruct images [265], imposing predictive consistency among augmented and interpolated data [266], and leveraging the nearest-neighbor classifier to distill class-instance relationships [267].”
BNS: Building Network Structures Dynamically for Continual Learning
Qin, Q., Hu, W., Han, P., Zhao, D., & Liu, B. (2021) · neural information processing systems · Cited by 26
Bridging Non Co-occurrence with Unlabeled In-the-wild Data for Incremental Object DetectionPreprint
Dong, N., Zhang, Y., Ding, M., & Lee, G. H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.15017 · Cited by 26 · cited 2×
“The idea of knowledge distillation is then introduced to other detection frameworks, such as CenterNet [522] (SID [338]), RetinaNet [267] (RILOD [255]), GFLV1 [258] (ERD [117]) and Faster R-CNN [369] (CIFRCN [155], Faster ILOD [337], DMC [506], BNC [99], IOD-ML [229]).”
MethodRamakrishnan, K., Panda, R., Fan, Q., Henning, J., Oliva, A., & Feris, R. S. (2020) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw50498.2020.00133 · Cited by 25 · cited 5×
“2, d), which is further validated by a theoretical analysis of the forgetting-generalization trade-off [350] and the PACBayes bound of generalization errors [330], [430].”
BackgroundMethodInfluentialA simple but strong baseline for online continual learning: Repeated Augmented RehearsalPreprint
Zhang, Y., Pfahringer, B., Frank, E., Bifet, A., Lim, N., & Jia, Y. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2209.13917 · Cited by 25
Routing Networks with Co-training for Continual LearningPreprint
Collier, M., Kokiopoulou, E., Gesmundo, A., & Berent, J. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2009.04381 · Cited by 25
Research works printed but not linked: 4
- Ramasesh, Effect of scale on catastrophic forgetting in neural networks
- Benzing, Unifying importance based regularisation methods for continual learning, 2372
- Miao, Continual learning with filter atom swapping
- Ardywibowo, VariGrow: Variational architecture growing for task-agnostic continual learning based on Bayesian novelty, 865
Cited by
Atakishiyev, S., Salameh, M., Yao, H., & Goebel, R. (2024) · IEEE Access · doi:10.1109/access.2024.3431437 · Cited by 323
Atakishiyev, S., Salameh, M., Yao, H., & Goebel, R. (2021) · arXiv · doi:10.48550/arxiv.2112.11561 · Cited by 336
Vision-language models for medical report generation and visual question answering: a review
Hartsock, I., & Rasool, G. (2024) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2024.1430984 · Cited by 229
A Comprehensive Survey of Convolutions in Deep Learning: Applications, Challenges, and Future Trends
Younesi, A., Ansari, M., Fazli, M., Ejlali, A., Shafique, M., & Henkel, J. (2024) · IEEE Access · doi:10.1109/access.2024.3376441 · Cited by 197
“Models must avoid catastrophic forgetting when presented with new classes or shifts in existing class definitions without revisiting historical data [391], [392], [393], [394], [395], [396], [397], [398], [399], [400].”
BackgroundVoyager: An Open-Ended Embodied Agent with Large Language Models
Wang, G., Xie, Y., Jiang, Y., Mandlekar, A., Xiao, C., Zhu, Y., Fan, L., & Anandkumar, A. (2023) · arXiv · doi:10.48550/arxiv.2305.16291 · Cited by 222
“[32] Liyuan Wang, Xingxing Zhang, Hang Su, and Jun Zhu.”
Background
Metadata
- Authors
- Wang, L., Zhang, X., Su, H., & Zhu, J.
- Affiliations
- Tsinghua University (CN)
- Venue
- IEEE Transactions on Pattern Analysis and Machine Intelligence
- Published
- 2024-08
- Volume
- 46
- Issue
- 8
- Pages
- 5362-5383
- ISSN
- 0162-8828 (ISSN-L)
- 2160-9292
- 1939-3539
- OpenAlex
- W4392173735
- Cited by
- 1,113
- References
- 313
- Citations by year
- 2026: 437 · 2025: 471 · 2024: 170
- FWCI
- 158.75
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- IEEE Transactions on Pattern Analysis and Machine Intelligence (publishedVersion)
- PubMed (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Natural Science Foundation of China (62350080, 62076147, 62106123, U19A2081)
- BNRist (BNR2022RC01006)
- Tsinghua Institute for Guo Qiang
- Tsinghua-OPPO Joint Research Center for Future Terminal Technology
- High Performance Computing Center
- Tsinghua University
- Postdoctoral Fellowship Program of CPSF (GZB20230350)
- Shuimu Tsinghua Scholar
- New Cornerstone Science Foundation
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Physical Sciences › Computer Science › Artificial Intelligence › Data Stream Mining Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Reinforcement Learning in Robotics
- Keywords
- continual learning, catastrophic forgetting, stability-plasticity trade-off, resource efficiency, incremental learning, adaptive intelligent systems
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.