外部记忆论文趋势 2023–2026

主要结论

• 外部记忆论文从 2023 年的 21 篇增长到 2026 年的 56 篇，三年间扩大到约 2.7 倍。
• NeurIPS 是发外部记忆论文最多的会议，2023–2026 年共 46 篇，占四个会议总量的 35%。
• 交叉方向里涨得最快的是AI 智能体：同时属于两个主题的论文从 2023 年的 0 篇增至 2026 年的 9 篇。
• 在「长上下文与记忆」的 5 个主题里，它的 2026 年论文数排第 5；同组增长最快的检索记忆是 6.6 倍，外部记忆是 2.7 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 2026Efficient Memory Crystallization for Graph Learning under Non-Stationary Distribution Shifts
• ICML 2026A Queueing-Theoretic Framework for Stability Analysis of LLM Inference with KV Cache Memory Constraints
• ICLR 2026Pre-training Limited Memory Language Models with Internal and External Knowledge
• CVPR 2026Spatial-SAM Spatially Consistent 3D Electron Microscopy Segmentation with SDF Memory
• NeurIPS 2025Future Link Prediction Without Memory or Aggregation
• ICML 2025Agent Workflow Memory
• ICLR 2025BitStack: Any-Size Compression of Large Language Models in Variable Memory Environments
• CVPR 2025Breaking the Memory Barrier of Contrastive Loss via Tile-Based Strategy
• NeurIPS 2024Pipeline Parallelism with Controllable Memory
• ICML 2024Graph External Attention Enhanced Transformer
• ICLR 2024The Cost of Scaling Down Large Language Models: Reducing Model Size Affects Memory before In-context Learning
• CVPR 2024RTracker: Recoverable Tracking via PN Tree Structured Memory
• NeurIPS 2023Transformer as a hippocampal memory consolidation model based on NMDAR-inspired nonlinearity
• ICML 2023Memory-Based Meta-Learning on Non-Stationary Distributions
• ICLR 2023Mass-Editing Memory in a Transformer
• CVPR 2023Bilateral Memory Consolidation for Continual Learning
• NeurIPS 2026Memory Grafting: Scaling Language Model Pre-training via Offline Conditional Memory
• ICML 2026Generalized Linear Bandits with Memory
• ICLR 2026QKV Projections Require a Fraction of Their Memory
• CVPR 2026Captain Safari A World Engine with Pose-Aligned 3D Memory

全部外部记忆论文

共 130 篇，第 3 / 3 页

101. ICML 2024Larimar: Large Language Models with Episodic Memory Control
102. ICML 2024Memoria: Resolving Fateful Forgetting Problem through Human-Inspired Memory Architecture
103. ICML 2024StableSSM: Alleviating the Curse of Memory in State-space Models through Stable Reparameterization
104. ICML 2024Structured Inverse-Free Natural Gradient Descent: Memory-Efficient & Numerically-Stable KFAC
105. ICML 2024Think Before You Act: Decision Transformers with Working Memory
106. ICML 2024Towards Interpretable Deep Local Learning with Successive Gradient Reconciliation
107. NeurIPS 20244-bit Shampoo for Memory-Efficient Network Training
108. NeurIPS 2024Pipeline Parallelism with Controllable Memory
109. NeurIPS 2024Towards Exact Gradient-based Training on Analog In-memory Computing
110. CVPR 2023Bilateral Memory Consolidation for Continual Learning
111. CVPR 2023GradMA: A Gradient-Memory-Based Accelerated Federated Learning With Alleviated Catastrophic Forgetting
112. CVPR 2023Hierarchical Neural Memory Network for Low Latency Event Processing
113. CVPR 2023Memory-Friendly Scalable Super-Resolution via Rewinding Lottery Ticket Hypothesis
114. CVPR 2023NaQ: Leveraging Narrations As Queries To Supervise Episodic Memory
115. CVPR 2023Token Turing Machines
116. ICLR 2023A Model or 603 Exemplars: Towards Memory-Efficient Class-Incremental Learning
117. ICLR 2023GAMR: A Guided Attention Model for (visual) Reasoning
118. ICLR 2023Mass-Editing Memory in a Transformer
119. ICLR 2023Memory Gym: Partially Observable Challenges to Memory-Based Agents
120. ICLR 2023POPGym: Benchmarking Partially Observable Reinforcement Learning
121. ICLR 2023Replay Memory as An Empirical MDP: Combining Conservative Estimation with Experience Replay
122. ICLR 2023Short-Term Memory Convolutions
123. ICLR 2023Simplicial Hopfield networks
124. ICML 2023Few-bit Backward: Quantized Gradients of Activation Functions for Memory Footprint Reduction
125. ICML 2023Memory-Based Meta-Learning on Non-Stationary Distributions
126. NeurIPS 2023Hypothesis Selection with Memory Constraints
127. NeurIPS 2023Memory Efficient Optimizers with 4-bit States
128. NeurIPS 2023Sequential Memory with Temporal Predictive Coding
129. NeurIPS 2023Thrust: Adaptively Propels Large Language Models with External Knowledge
130. NeurIPS 2023Transformer as a hippocampal memory consolidation model based on NMDAR-inspired nonlinearity

常见问题

2023–2026 年四大会议一共有多少篇外部记忆论文？

共 130 篇，按年份分别是 2023 年 21 篇、2024 年 25 篇、2025 年 28 篇、2026 年 56 篇。

NeurIPS 2026 有多少篇外部记忆论文？

NeurIPS 2026 有 26 篇外部记忆论文，上一届是 12 篇。

ICLR 2026 有多少篇外部记忆论文？

ICLR 2026 有 13 篇外部记忆论文，上一届是 10 篇。

ICML 2026 有多少篇外部记忆论文？

ICML 2026 有 13 篇外部记忆论文，上一届是 3 篇。

CVPR 2026 有多少篇外部记忆论文？

CVPR 2026 有 4 篇外部记忆论文，上一届是 3 篇。

哪个会议的外部记忆论文最多？

NeurIPS，2023–2026 年共 46 篇。