循环记忆论文趋势 2023–2026

主要结论

• 循环记忆论文从 2023 年的 45 篇增长到 2026 年的 172 篇，三年间扩大到约 3.8 倍。
• NeurIPS 是发循环记忆论文最多的会议，2023–2026 年共 186 篇，占四个会议总量的 45%。
• 交叉方向里涨得最快的是多模态大模型：同时属于两个主题的论文从 2023 年的 0 篇增至 2026 年的 15 篇。
• 在「长上下文与记忆」的 5 个主题里，它的 2026 年论文数排第 2；同组增长最快的检索记忆是 6.6 倍，循环记忆是 3.8 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 2026Approximate Memory Suffices for Universal Approximation with State Space Models
• ICML 2026An Algebraic View of the Expressivity of Recurrent Language Models
• ICLR 2026Exploring State-Space Models for Data-Specific Neural Representations
• CVPR 2026Coupling Liquid Time-Constant Encoders with Modern Hopfield Memory
• NeurIPS 2025Bridging Expressivity and Scalability with Adaptive Unitary SSMs
• ICML 2025Oracle-MoE: Locality-preserving Routing in the Oracle Space for Memory-constrained Large Language Model Inference
• ICLR 2025Autocorrelation Matters: Understanding the Role of Initialization Schemes for State Space Models
• CVPR 2025Breaking the Memory Barrier of Contrastive Loss via Tile-Based Strategy
• NeurIPS 2024Attractor Memory for Long-Term Time Series Forecasting: A Chaos Perspective
• ICML 2024Hidden Traveling Waves bind Working Memory Variables in Recurrent Neural Networks
• ICLR 2024A 2-Dimensional State Space Layer for Spatial Inductive Bias
• CVPR 2024GenFlow: Generalizable Recurrent Flow for 6D Pose Refinement of Novel Objects
• NeurIPS 2023Hypothesis Selection with Memory Constraints
• ICML 2023A Picture of the Space of Typical Learnable Tasks
• ICLR 2023Associative Memory Augmented Asynchronous Spatiotemporal Representation Learning for Event-based Perception
• CVPR 2023Bilateral Memory Consolidation for Continual Learning
• NeurIPS 2026Commutator Memory: Sparse, Path-Local Reading and Steering in Language Models
• ICML 2026Do Transformers Need Three Projections? Systematic Study of QKV Variants
• ICLR 2026On the Expressiveness of State Space Models via Temporal Logics
• CVPR 2026DREAM Document Recognition with Explicit Adaptive Memory

全部循环记忆论文

共 186 篇，第 2 / 4 页

51. NeurIPS 2026PatchKV: Weight Space Compensation of KV Cache
52. NeurIPS 2026PDE-SSM: A Spectral State Space Approach to Spatial Mixing in Diffusion Transformers
53. NeurIPS 2026PREPING: Building Agent Memory without Tasks
54. NeurIPS 2026Prospective Coding Improves Learning in Deep Continuous-Time Recurrent Networks
55. NeurIPS 2026Raven: High-Recall Sequence Modeling via Sparse Memory Routing
56. NeurIPS 2026RecMem: Recurrent Memory Compression for Long-Sequence Recommendation
57. NeurIPS 2026Remember the Decision, Not the Description: A Rate-Distortion Framework for Agent Memory
58. NeurIPS 2026Rethinking State Tracking in Recurrent Models Through Error Control Dynamics
59. NeurIPS 2026Rethinking the State Update Gate for Long-Sequence Recurrent 3D Reconstruction
60. NeurIPS 2026Selectivity Makes State-Space Models Universal Sequence Approximators
61. NeurIPS 2026Skipping Domain Shifts: Domain Memory Retention Enhanced Hyperspectral Single-Source Domain Generalization
62. NeurIPS 2026SLVMBench: Skill Learning from Video Memory
63. NeurIPS 2026Space Group Conditional Flow Matching
64. NeurIPS 2026Sparse Delta Memory: Scaling the State of Linear RNNs through Sparsity
65. NeurIPS 2026Spectral Quantum Memory for Implicit Neural Representations
66. NeurIPS 2026SpikeSSL: A Universal Spike Inference Framework with Dynamics-Informed State-Space Layers
67. NeurIPS 2026State Augmented Flows
68. NeurIPS 2026Statistical Mixing Guarantees for Contractive Echo State Networks
69. NeurIPS 2026StreamPhy: Streaming Inference of High-Dimensional Physical Dynamics via State Space Models
70. NeurIPS 2026Structured Sparse Memory for Recurrent Reasoning
71. NeurIPS 2026Swimba: Switch Mamba Model Scales State Space Models
72. NeurIPS 2026The State-Prediction Separation Hypothesis
73. NeurIPS 2026TIDES: Implicit Time-Awareness in Selective State Space Models
74. NeurIPS 2025Bridging Expressivity and Scalability with Adaptive Unitary SSMs
75. NeurIPS 2025Computation and Memory-Efficient Model Compression with Gradient Reweighting
76. NeurIPS 2025Cost-Efficient LLM Training with Lifetime-Aware Tensor Offloading via GPUDirect Storage
77. NeurIPS 2025Decoder-Hybrid-Decoder Architecture for Efficient Reasoning with Long Generation
78. NeurIPS 2025DeltaFormer: Unlock the state space of Transformer
79. NeurIPS 2025DeltaProduct: Improving State-Tracking in Linear RNNs via Householder Products
80. NeurIPS 2025DERD-Net: Learning Depth from Event-based Ray Densities
81. NeurIPS 2025DyG-Mamba: Continuous State Space Modeling on Dynamic Graphs
82. NeurIPS 2025DyMoDreamer: World Modeling with Dynamic Modulation
83. NeurIPS 2025Efficient Allocation of Working Memory Resource for Utility Maximization in Humans and Recurrent Neural Networks
84. NeurIPS 2025Error Forcing in Recurrent Neural Networks
85. NeurIPS 2025Exponential Dynamic Energy Network for High Capacity Sequence Memory
86. NeurIPS 2025Fixed-Point RNNs: Interpolating from Diagonal to Dense
87. NeurIPS 2025FLAME: Fast Long-context Adaptive Memory for Event-based Vision
88. NeurIPS 2025Flow Equivariant Recurrent Neural Networks
89. NeurIPS 2025FRN: Fractal-Based Recursive Spectral Reconstruction Network
90. NeurIPS 2025Hankel Singular Value Regularization for Highly Compressible State Space Models
91. NeurIPS 2025Hardware-aligned Hierarchical Sparse Attention for Efficient Long-term Memory Access
92. NeurIPS 2025Inference of Whole Brain Electrophysiological Networks Through Multimodal Integration of Simultaneous Scalp and Intracranial EEG
93. NeurIPS 2025Inference-time Alignment in Continuous Space
94. NeurIPS 2025Irrational Complex Rotations Empower Low-bit Optimizers
95. NeurIPS 2025Kinaema: a recurrent sequence model for memory and pose in motion
96. NeurIPS 2025L$^2$M: Mutual Information Scaling Law for Long-Context Language Modeling
97. NeurIPS 2025Learning Memory-Enhanced Improvement Heuristics for Flexible Job Shop Scheduling
98. NeurIPS 2025Mamba Modulation: On the Length Generalization of Mamba Models
99. NeurIPS 2025Measuring and Controlling Solution Degeneracy across Task-Trained Recurrent Neural Networks
100. NeurIPS 2025Memo: Training Memory-Efficient Embodied Agents with Reinforcement Learning

常见问题

2023–2026 年四大会议一共有多少篇循环记忆论文？

共 416 篇，按年份分别是 2023 年 45 篇、2024 年 82 篇、2025 年 117 篇、2026 年 172 篇。

NeurIPS 2026 有多少篇循环记忆论文？

NeurIPS 2026 有 73 篇循环记忆论文，上一届是 61 篇。

ICLR 2026 有多少篇循环记忆论文？

ICLR 2026 有 35 篇循环记忆论文，上一届是 20 篇。

ICML 2026 有多少篇循环记忆论文？

ICML 2026 有 43 篇循环记忆论文，上一届是 16 篇。

CVPR 2026 有多少篇循环记忆论文？

CVPR 2026 有 21 篇循环记忆论文，上一届是 20 篇。

哪个会议的循环记忆论文最多？

NeurIPS，2023–2026 年共 186 篇。