持续学习论文趋势 2023–2026

主要结论

• 持续学习论文从 2023 年的 77 篇增长到 2026 年的 131 篇，三年间扩大到约 1.7 倍。
• NeurIPS 是发持续学习论文最多的会议，2023–2026 年共 121 篇，占四个会议总量的 32%。
• 交叉方向里涨得最快的是低秩适配：同时属于两个主题的论文从 2023 年的 1 篇增至 2026 年的 16 篇。
• 在「元学习与迁移」的 5 个主题里，它的 2026 年论文数排第 5；同组增长最快的基础模型适配是 3.8 倍，持续学习是 1.7 倍。

代表论文

按年份和会议均匀挑选，方便快速看到每一届的样貌

• NeurIPS 2026Bayes-pFCL:Bayesian Personalized Federated Continual Learning
• ICML 2026Beyond Buffer Limits: Energy-Based Data Reassembly for Continual Learning
• ICLR 2026FlyPrompt: Brain-Inspired Random-Expanded Routing with Temporal-Ensemble Experts for General Continual Learning
• CVPR 2026A Faster Path to Continual Learning
• NeurIPS 2025AnaCP: Toward Upper-Bound Continual Learning via Analytic Contrastive Projection
• ICML 2025BECAME: Bayesian Continual Learning with Adaptive Model Merging
• ICLR 2025Active Learning for Continual Learning: Keeping the Past Alive in the Present
• CVPR 2025AdaptCMVC: Robust Adaption to Incremental Views in Continual Multi-view Clustering
• NeurIPS 2024CoIN: A Benchmark of Continual Instruction Tuning for Multimodel Large Language Models
• ICML 2024An Effective Dynamic Gradient Calibration Method for Continual Learning
• ICLR 2024A Probabilistic Framework for Modular Continual Learning
• CVPR 2024Defense without Forgetting: Continual Adversarial Defense with Anisotropic & Isotropic Pseudo Replay
• NeurIPS 2023Aging with GRACE: Lifelong Model Editing with Discrete Key-Value Adaptors
• ICML 2023Constant Matters: Fine-grained Error Bound on Differentially Private Continual Observation
• ICLR 2023Artificial Neuronal Ensembles with Learned Context Dependent Gating
• CVPR 2023Achieving a Better Stability-Plasticity Trade-Off via Auxiliary Networks in Continual Learning
• NeurIPS 2026Beyond Exemplar Selection: Value-Aware Memory Allocation in Replay-Based Continual Learning
• ICML 2026BTSP-CAM: A Brain-Inspired Geometric Memory for Class-Incremental Learning
• ICLR 2026IDER: IDEMPOTENT EXPERIENCE REPLAY FOR RELIABLE CONTINUAL LEARNING
• CVPR 2026AdaPrior Bayesian-Inspired Adaptive Prior Correction for Long-Tailed Continual Learning

全部持续学习论文

共 104 篇，第 2 / 3 页

51. ICML 2025Demystifying Catastrophic Forgetting in Two-Stage Incremental Object Detector
52. ICML 2025Discovering a Zero (Zero-Vector Class of Machine Learning)
53. ICML 2025FedSSI: Rehearsal-Free Continual Federated Learning with Synergistic Synaptic Intelligence
54. ICML 2025LADA: Scalable Label-Specific CLIP Adapter for Continual Learning
55. ICML 2025Large Continual Instruction Assistant
56. ICML 2025Learning without Isolation: Pathway Protection for Continual Learning
57. ICML 2025Measuring Representational Shifts in Continual Learning: A Linear Transformation Perspective
58. ICML 2025Mitigating Plasticity Loss in Continual Reinforcement Learning by Reducing Churn
59. ICML 2025Model Uncertainty Quantification by Conformal Prediction in Continual Learning
60. ICML 2025Online Curvature-Aware Replay: Leveraging $\mathbf{2^{nd}}$ Order Information for Online Continual Learning
61. ICML 2025Optimal Task Order for Continual Learning of Multiple Tasks
62. ICML 2025Overtrained Language Models Are Harder to Fine-Tune
63. ICML 2025Position: Lifetime tuning is incompatible with continual reinforcement learning
64. ICML 2025Predicting the Susceptibility of Examples to Catastrophic Forgetting
65. ICML 2025Rethinking the Stability-Plasticity Trade-off in Continual Learning from an Architectural Perspective
66. ICML 2025SEFE: Superficial and Essential Forgetting Eliminator for Multimodal Continual Instruction Tuning
67. ICML 2025The Importance of Being Lazy: Scaling Limits of Continual Learning
68. ICML 2025Towards Understanding Catastrophic Forgetting in Two-layer Convolutional Neural Networks
69. ICML 2025Understanding the Forgetting of (Replay-based) Continual Learning via Feature Learning: Angle Matters
70. ICML 2025Unlocking the Power of Rehearsal in Continual Learning: A Theoretical Perspective
71. ICML 2025Where is the Truth? The Risk of Getting Confounded in a Continual World
72. ICML 2025ZeroFlow: Overcoming Catastrophic Forgetting is Easier than You Think
73. NeurIPS 2025AnaCP: Toward Upper-Bound Continual Learning via Analytic Contrastive Projection
74. NeurIPS 2025Are Greedy Task Orderings Better Than Random in Continual Linear Regression?
75. NeurIPS 2025Auto-Compressing Networks
76. NeurIPS 2025C-NAV: Towards Self-Evolving Continual Object Navigation in Open World
77. NeurIPS 2025Compact Memory for Continual Logistic Regression
78. NeurIPS 2025Continual Knowledge Adaptation for Reinforcement Learning
79. NeurIPS 2025Continual Multimodal Contrastive Learning
80. NeurIPS 2025Continuous Subspace Optimization for Continual Learning
81. NeurIPS 2025Contrastive Consolidation of Top-Down Modulations Achieves Sparsely Supervised Continual Learning
82. NeurIPS 2025Decentralized Dynamic Cooperation of Personalized Models for Federated Continual Learning
83. NeurIPS 2025Demystifying Language Model Forgetting with Low-rank Example Associations
84. NeurIPS 2025Dual-Space Semantic Synergy Distillation for Continual Learning of Unlabeled Streams
85. NeurIPS 2025Dynamic Siamese Expansion Framework for Improving Robustness in Online Continual Learning
86. NeurIPS 2025Federated Continual Learning via Orchestrating Multi-Scale Expertise
87. NeurIPS 2025Gated Integration of Low-Rank Adaptation for Continual Learning of Large Language Models
88. NeurIPS 2025Gradient-Guided Epsilon Constraint Method for Online Continual Learning
89. NeurIPS 2025Hippocampal-like Sequential Editing for Continual Knowledge Updates in Large Language Models
90. NeurIPS 2025Hybrid Re-matching for Continual Learning with Parameter-Efficient Tuning
91. NeurIPS 2025Memory-Integrated Reconfigurable Adapters: A Unified Framework for Settings with Multiple Tasks
92. NeurIPS 2025Merging on the Fly Without Retraining: A Sequential Approach to Scalable Continual Model Merging
93. NeurIPS 2025MINGLE: Mixture of Null-Space Gated Low-Rank Experts for Test-Time Continual Model Merging
94. NeurIPS 2025Mitigating Intra- and Inter-modal Forgetting in Continual Learning of Unified Multimodal Models
95. NeurIPS 2025Online Functional Tensor Decomposition via Continual Learning for Streaming Data Completion
96. NeurIPS 2025Optimal Rates in Continual Linear Regression via Increasing Regularization
97. NeurIPS 2025OVS Meets Continual Learning: Towards Sustainable Open-Vocabulary Segmentation
98. NeurIPS 2025REP: Resource-Efficient Prompting for Rehearsal-Free Continual Learning
99. NeurIPS 2025Self-Evolving Pseudo-Rehearsal for Catastrophic Forgetting with Task Similarity in LLMs
100. NeurIPS 2025SPICED: A Synaptic Homeostasis-Inspired Framework for Unsupervised Continual EEG Decoding

常见问题

2023–2026 年四大会议一共有多少篇持续学习论文？

共 381 篇，按年份分别是 2023 年 77 篇、2024 年 69 篇、2025 年 104 篇、2026 年 131 篇。

NeurIPS 2026 有多少篇持续学习论文？

NeurIPS 2026 有 39 篇持续学习论文，上一届是 32 篇。

ICLR 2026 有多少篇持续学习论文？

ICLR 2026 有 28 篇持续学习论文，上一届是 23 篇。

ICML 2026 有多少篇持续学习论文？

ICML 2026 有 42 篇持续学习论文，上一届是 29 篇。

CVPR 2026 有多少篇持续学习论文？

CVPR 2026 有 22 篇持续学习论文，上一届是 20 篇。

哪个会议的持续学习论文最多？

NeurIPS，2023–2026 年共 121 篇。