流式推理论文趋势 2023–2026

主要结论

• 流式推理论文从 2023 年的 13 篇增长到 2026 年的 41 篇，三年间扩大到约 3.2 倍。
• NeurIPS 是发流式推理论文最多的会议，2023–2026 年共 35 篇，占四个会议总量的 39%。
• 交叉方向里涨得最快的是三维视觉：同时属于两个主题的论文从 2023 年的 0 篇增至 2026 年的 10 篇。
• 在「边缘与物联网」的 5 个主题里，它的 2026 年论文数排第 3；同组增长最快的传感融合是 3.8 倍，流式推理是 3.2 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 2026MIST: Reliable Streaming Decision Trees for Online Class-Incremental Learning via McDiarmid Bound
• ICML 2026Estimating Correlation Clustering Cost in Node-Arrival Stream
• ICLR 2026Near-Optimal Online Deployment and Routing for Streaming LLMs
• CVPR 2026Color When It Counts Grayscale-Guided Online Triggering for Always-On Streaming
• NeurIPS 2025Learning-Augmented Streaming Algorithms for Correlation Clustering
• ICML 2025Deep Streaming View Clustering
• ICLR 2025Learning-Augmented Frequent Directions
• CVPR 2025Learning from Streaming Video with Orthogonal Gradients
• NeurIPS 2024Stochastic Optimization Algorithms for Instrumental Variable Regression with Streaming Data
• ICML 2024Fast White-Box Adversarial Streaming Without a Random Oracle
• ICLR 2024Addressing Loss of Plasticity and Catastrophic Forgetting in Continual Learning
• CVPR 2024Streaming Dense Video Captioning
• NeurIPS 2023Streaming PCA for Markovian Data
• ICML 2023Lookahead When It Matters: Adaptive Non-causal Transformers for Streaming Neural Transducers
• ICLR 2023Continual Transformers: Redundancy-Free Attention for Online Inference
• CVPR 2023System-Status-Aware Adaptive Network for Online Streaming Video Understanding
• NeurIPS 2026QueryStop: Dynamic Stop Signals for Efficient Streaming Inference
• ICML 2026SFedPO: Streaming Federated Learning with a Prediction Oracle under Temporal Shifts
• CVPR 2026Event Stream Filtering via Probability Flux Estimation
• NeurIPS 2025Online Functional Tensor Decomposition via Continual Learning for Streaming Data Completion

全部流式推理论文

共 15 篇，第 1 / 1 页

1. NeurIPS 2026Dynamics of the Transformer Residual Stream: Coupling Spectral Geometry to Network Topology
2. NeurIPS 2026FrameVGGT: Coherence-Preserving Memory for Bounded Streaming Geometry
3. NeurIPS 2026GHOST: Geometry-Hierarchical Online Streaming Token Eviction for Efficient 3D Reconstruction
4. NeurIPS 2026HIST3R: Rectified State Decomposition from History for Streaming 3D Reconstruction
5. NeurIPS 2026IGGT4D: Streaming 4D Instance-Grounded Geometry Transformer
6. NeurIPS 2026Inference and Uncertainty Quantification for Streaming $r$-PCA
7. NeurIPS 2026MIST: Reliable Streaming Decision Trees for Online Class-Incremental Learning via McDiarmid Bound
8. NeurIPS 2026Online Directional Regression for Streaming Sufficient Dimension Reduction
9. NeurIPS 2026PF-SGS: Pose-Free Streaming 3D Gaussian Splatting for Large-Scale Scene Reconstruction
10. NeurIPS 2026QueryStop: Dynamic Stop Signals for Efficient Streaming Inference
11. NeurIPS 2026SoccerNarrate: Event-Grounded Streaming Soccer Commentary with Macro-Window Preference Alignment
12. NeurIPS 2026Stable3R: Streaming 3D Reconstruction with Stable Geometric References
13. NeurIPS 2026Streaming Interventions: Can Video LLMs Correct Mistakes as They Occur?
14. NeurIPS 2026StreamPhy: Streaming Inference of High-Dimensional Physical Dynamics via State Space Models
15. NeurIPS 2026What Should a Streaming Video Model Remember?

常见问题

2023–2026 年四大会议一共有多少篇流式推理论文？

共 90 篇，按年份分别是 2023 年 13 篇、2024 年 19 篇、2025 年 17 篇、2026 年 41 篇。

NeurIPS 2026 有多少篇流式推理论文？

NeurIPS 2026 有 15 篇流式推理论文，上一届是 8 篇。

ICLR 2026 有多少篇流式推理论文？

ICLR 2026 有 1 篇流式推理论文，上一届是 2 篇。

ICML 2026 有多少篇流式推理论文？

ICML 2026 有 10 篇流式推理论文，上一届是 4 篇。

CVPR 2026 有多少篇流式推理论文？

CVPR 2026 有 15 篇流式推理论文，上一届是 3 篇。

哪个会议的流式推理论文最多？

NeurIPS，2023–2026 年共 35 篇。