算子融合论文趋势 2023–2026

主要结论

• 算子融合论文从 2023 年的 51 篇降到 2026 年的 48 篇，热度在回落。
• NeurIPS 是发算子融合论文最多的会议，2023–2026 年共 73 篇，占四个会议总量的 33%。
• 交叉方向里涨得最快的是传感融合：同时属于两个主题的论文从 2023 年的 2 篇增至 2026 年的 12 篇。
• 在「硬件与编译」的 5 个主题里，它的 2026 年论文数排第 3；同组增长最快的端侧推理是 4.0 倍，算子融合是 0.9 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 2026Beyond Adjacent Layers: Graph-Guided Layer Fusion for Compressing Large Language Models
• ICML 2026Asymmetric conformal prediction with penalized kernel sum-of-squares
• ICLR 2026Probabilistic Kernel Function for Fast Angle Testing
• CVPR 2026Bridging RGB and Hematoxylin Components An Interleaved Guidance and Fusion
• NeurIPS 2025Beyond Benign Overfitting in Nadaraya-Watson Interpolators
• ICML 2025Equivariant Neural Tangent Kernels
• ICLR 2025DLEFT-MKC: Dynamic Late Fusion Multiple Kernel Clustering with Robust Tensor Learning via Min-Max Optimization
• CVPR 2025Parameterized Blur Kernel Prior Learning for Local Motion Deblurring
• NeurIPS 2024Efficient Convex Algorithms for Universal Kernel Learning
• ICML 2024Characterizing Overfitting in Kernel Ridgeless Regression Through the Eigenspectrum
• ICLR 2024Convolutional Deep Kernel Machines
• CVPR 2024Bi-SSC: Geometric-Semantic Bidirectional Fusion for Camera-based 3D Semantic Scene Completion
• NeurIPS 2023CosNet: A Generalized Spectral Kernel Network
• ICML 2023A Kernel Stein Test of Goodness of Fit for Sequential Models
• ICLR 2023A Kernel Perspective of Skip Connections in Convolutional Networks
• CVPR 2023Compacting Binary Neural Networks by Sparse Kernel Selection
• NeurIPS 2026Dynamical Adapter Fusion: Constructing A Global Adapter for Pre-Trained Model-based Class-Incremental Learning
• ICML 2026Even Faster Kernel Matrix Linear Algebra via Density Estimation
• ICLR 2026LaplacianFormer:Rethinking Linear Attention with Laplacian Kernel
• CVPR 2026CryoKRAQEN Kernel-Regularized Annealing for Quantized Embedding Networks in Cryo-EM Heterogeneous

全部算子融合论文

共 66 篇，第 1 / 2 页

1. CVPR 2024Bi-SSC: Geometric-Semantic Bidirectional Fusion for Camera-based 3D Semantic Scene Completion
2. CVPR 2024Data-Efficient Multimodal Fusion on a Single GPU
3. CVPR 2024Embracing Unimodal Aleatoric Uncertainty for Robust Multimodal Fusion
4. CVPR 2024KPConvX: Modernizing Kernel Point Convolution with Kernel Attention
5. CVPR 2024PeLK: Parameter-efficient Large Kernel ConvNets with Peripheral Convolution
6. CVPR 2024Poly Kernel Inception Network for Remote Sensing Detection
7. CVPR 2024Robust Depth Enhancement via Polarization Prompt Fusion Tuning
8. CVPR 2024SecondPose: SE(3)-Consistent Dual-Stream Feature Fusion for Category-Level Pose Estimation
9. CVPR 2024SFOD: Spiking Fusion Object Detector
10. ICLR 2024An Agnostic View on the Cost of Overfitting in (Kernel) Ridge Regression
11. ICLR 2024Convolutional Deep Kernel Machines
12. ICLR 2024Counterfactual Density Estimation using Kernel Stein Discrepancies
13. ICLR 2024Explaining Kernel Clustering via Decision Trees
14. ICLR 2024Fusing Models with Complementary Expertise
15. ICLR 2024lpNTK: Better Generalisation with Less Data via Sample Interaction During Learning
16. ICLR 2024Posterior Sampling Based on Gradient Flows of the MMD with Negative Distance Kernel
17. ICLR 2024Scalable Neural Network Kernels
18. ICLR 2024Smooth ECE: Principled Reliability Diagrams via Kernel Smoothing
19. ICLR 2024The optimality of kernel classifiers in Sobolev space
20. ICLR 2024Towards Meta-Pruning via Optimal Transport
21. ICLR 2024Transformer Fusion with Optimal Transport
22. ICML 2024Amortized Variational Deep Kernel Learning
23. ICML 2024Bring Your Own (Non-Robust) Algorithm to Solve Robust MDPs by Estimating The Worst Kernel
24. ICML 2024Characterizing Overfitting in Kernel Ridgeless Regression Through the Eigenspectrum
25. ICML 2024Copula-Nested Spectral Kernel Network
26. ICML 2024Deep Fusion: Efficient Network Training via Pre-trained Initializations
27. ICML 2024DeepPolar: Inventing Nonlinear Large-Kernel Polar Codes via Deep Learning
28. ICML 2024Dissecting Multimodality in VideoQA Transformer Models by Impairing Modality Fusion
29. ICML 2024Generalization in Kernel Regression Under Realistic Assumptions
30. ICML 2024High-Dimensional Kernel Methods under Covariate Shift: Data-Dependent Implicit Regularization
31. ICML 2024Is Kernel Prediction More Powerful than Gating in Convolutional Neural Networks?
32. ICML 2024Kernel Debiased Plug-in Estimation: Simultaneous, Automated Debiasing without Influence Functions for Many Target Parameters
33. ICML 2024Kernel Semi-Implicit Variational Inference
34. ICML 2024Kernel-Based Evaluation of Conditional Biological Sequence Models
35. ICML 2024Mean-field Analysis on Two-layer Neural Networks from a Kernel Perspective
36. ICML 2024Neural-Kernel Conditional Mean Embeddings
37. ICML 2024On the Consistency of Kernel Methods with Dependent Observations
38. ICML 2024Optimal Kernel Quantile Learning with Random Features
39. ICML 2024Position: Opportunities Exist for Machine Learning in Magnetic Fusion Energy
40. ICML 2024Predictive Dynamic Fusion
41. ICML 2024Quasi-Monte Carlo Features for Kernel Approximation
42. ICML 2024Revealing the Dark Secrets of Extremely Large Kernel ConvNets on Robustness
43. ICML 2024Reward-Free Kernel-Based Reinforcement Learning
44. ICML 2024Sampling in Unit Time with Kernel Fisher-Rao Flow
45. ICML 2024Scalable Multiple Kernel Clustering: Learning Clustering Structure from Expectation
46. ICML 2024Sequential Kernel Goodness-of-fit Testing
47. NeurIPS 2024A Comprehensive Analysis on the Learning Curve in Kernel Ridge Regression
48. NeurIPS 2024A generalized neural tangent kernel for surrogate gradient learning
49. NeurIPS 2024A Kernel Perspective on Distillation-based Collaborative Learning
50. NeurIPS 2024ACFun: Abstract-Concrete Fusion Facial Stylization

常见问题

2023–2026 年四大会议一共有多少篇算子融合论文？

共 221 篇，按年份分别是 2023 年 51 篇、2024 年 66 篇、2025 年 56 篇、2026 年 48 篇。

NeurIPS 2026 有多少篇算子融合论文？

NeurIPS 2026 有 14 篇算子融合论文，上一届是 25 篇。

ICLR 2026 有多少篇算子融合论文？

ICLR 2026 有 9 篇算子融合论文，上一届是 8 篇。

ICML 2026 有多少篇算子融合论文？

ICML 2026 有 13 篇算子融合论文，上一届是 14 篇。

CVPR 2026 有多少篇算子融合论文？

CVPR 2026 有 12 篇算子融合论文，上一届是 9 篇。

哪个会议的算子融合论文最多？

NeurIPS，2023–2026 年共 73 篇。