高效训练论文趋势 2023–2026
Scaling law、MoE、蒸馏与训练效率
统计范围：NeurIPS、ICLR、ICML、CVPR 四个会议，2023–2026 年，共 1,765 篇被归入该主题的论文。
主要结论
- 高效训练论文从 2023 年的 346 篇增长到 2026 年的 603 篇，三年间扩大到约 1.7 倍。
- NeurIPS 是发高效训练论文最多的会议，2023–2026 年共 499 篇，占四个会议总量的 28%。
- 交叉方向里涨得最快的是深度学习 · large language models：同时属于两个主题的论文从 2023 年的 6 篇增至 2026 年的 50 篇。
- 在「效率与系统」的 5 个主题里，它的 2026 年论文数排第 3；同组增长最快的硬件协同是 4.4 倍，高效训练是 1.7 倍。
各会议论文数
按年份和会议统计
| 年份 | NeurIPS | ICLR | ICML | CVPR | 合计 |
|---|---|---|---|---|---|
| 2023 | 112 | 67 | 83 | 84 | 346 |
| 2024 | 20 | 78 | 122 | 80 | 300 |
| 2025 | 190 | 126 | 118 | 82 | 516 |
| 2026 | 177 | 148 | 169 | 109 | 603 |
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 2026Adaptive Residual Quantization for Memory-Efficient Temporal Action Segmentation
- ICML 2026A Geometry-Aware Efficient Algorithm for Compositional Entropic Risk Minimization
- ICLR 2026An Efficient SE(p)-Invariant Transport Metric Driven by Polar Transport Discrepancy-based Representation
- CVPR 2026A Bit is All You Need Efficient Video Capture via
- NeurIPS 20254DGCPro: Efficient Hierarchical 4D Gaussian Compression for Progressive Volumetric Video Streaming
- ICML 2025A Sample Efficient Conditional Independence Test in the Presence of Discretization
- ICLR 2025Breaking Class Barriers: Efficient Dataset Distillation via Inter-Class Feature Compensator
- CVPR 2025Accurate Scene Text Recognition with Efficient Model Scaling and Cloze Self-Distillation
- NeurIPS 2024$\texttt{dattri}$: A Library for Efficient Data Attribution
- ICML 2024A fast algorithm to simulate nonlinear resistive networks
- ICLR 2024Dictionary Contrastive Learning for Efficient Local Supervision without Auxiliary Networks
- CVPR 2024Deep-TROJ: An Inference Stage Trojan Insertion Algorithm through Efficient Weight Replacement Attack
- NeurIPS 2023Alternating Updates for Efficient Transformers
- ICML 2023Auxiliary Modality Learning with Generalized Curriculum Distillation
- ICLR 2023A Model or 603 Exemplars: Towards Memory-Efficient Class-Incremental Learning
- CVPR 2023A Rotation-Translation-Decoupled Solution for Robust and Efficient Visual-Inertial Initialization
- NeurIPS 2026Advanced Routing as Regularization Allocation for Efficient Diffusion Transformer Training
- ICML 2026Active Learning with Foundation Model Priors: Efficient Learning under Class Imbalance
- ICLR 2026AQER: A Scalable and Efficient Data Loader for Digital Quantum Computers
- CVPR 2026Adaptive Bayesian Early-Exit Networks for Efficient Non-Transferable Learning
全部高效训练论文
共 492 篇，第 5 / 10 页
- ICML 2025Efficient Heterogeneity-Aware Federated Active Data Selection
- ICML 2025Efficient Logit-based Knowledge Distillation of Deep Spiking Neural Networks for Full-Range Timestep Deployment
- ICML 2025Efficient Long Context Fine-tuning with Chunk Flow
- ICML 2025Efficient Network Automatic Relevance Determination
- ICML 2025Efficient Optimization with Orthogonality Constraint: a Randomized Riemannian Submanifold Method
- ICML 2025Efficient Parallel Training Methods for Spiking Neural Networks with Constant Time Complexity
- ICML 2025Efficient Quantification of Multimodal Interaction at Sample Level
- ICML 2025Efficient Robust Conformal Prediction via Lipschitz-Bounded Networks
- ICML 2025Efficient Time Series Processing for Transformers and State-Space Models through Token Merging
- ICML 2025Equivariant Polynomial Functional Networks
- ICML 2025Fast Inference with Kronecker-Sparse Matrices
- ICML 2025FastCAV: Efficient Computation of Concept Activation Vectors for Explaining Deep Neural Networks
- ICML 2025FedOne: Query-Efficient Federated Learning for Black-box Discrete Prompt Learning
- ICML 2025FedSMU: Communication-Efficient and Generalization-Enhanced Federated Learning through Symbolic Model Updates
- ICML 2025FisherSFT: Data-Efficient Supervised Fine-Tuning of Language Models Using Information Gain
- ICML 2025Flexible and Efficient Grammar-Constrained Decoding
- ICML 2025Flexible, Efficient, and Stable Adversarial Attacks on Machine Unlearning
- ICML 2025FloE: On-the-Fly MoE Inference on Memory-constrained GPU
- ICML 2025FRUGAL: Memory-Efficient Optimization by Reducing State Overhead for Scalable Training
- ICML 2025FSL-SAGE: Accelerating Federated Split Learning via Smashed Activation Gradient Estimation
- ICML 2025GPTAQ: Efficient Finetuning-Free Quantization for Asymmetric Calibration
- ICML 2025H-Tuning: Toward Low-Cost and Efficient ECG-based Cardiovascular Disease Detection with Pre-Trained Models
- ICML 2025HiRemate: Hierarchical Approach for Efficient Re-materialization of Neural Networks
- ICML 2025How to Train Your Multi-Exit Model? Analyzing the Impact of Training Strategies
- ICML 2025IL-SOAR : Imitation Learning with Soft Optimistic Actor cRitic
- ICML 2025Improved and Oracle-Efficient Online $\ell_1$-Multicalibration
- ICML 2025Improving Continual Learning Performance and Efficiency with Auxiliary Classifiers
- ICML 2025Info-Coevolution: An Efficient Framework for Data Model Coevolution
- ICML 2025Joint MoE Scaling Laws: Mixture of Experts Can Be Memory Efficient
- ICML 2025KVTuner: Sensitivity-Aware Layer-Wise Mixed-Precision KV Cache Quantization for Efficient and Nearly Lossless LLM Inference
- ICML 2025LaCache: Ladder-Shaped KV Caching for Efficient Long-Context Modeling of Large Language Models
- ICML 2025Learning from Loss Landscape: Generalizable Mixed-Precision Quantization via Adaptive Sharpness-Aware Gradient Aligning
- ICML 2025Learning Soft Sparse Shapes for Efficient Time-Series Classification
- ICML 2025LensLLM: Unveiling Fine-Tuning Dynamics for LLM Selection
- ICML 2025Leveraging Sparsity for Sample-Efficient Preference Learning: A Theoretical Perspective
- ICML 2025Memory Layers at Scale
- ICML 2025MERGE$^3$: Efficient Evolutionary Merging on Consumer-grade GPUs
- ICML 2025Mixture of Hidden-Dimensions: Not All Hidden-States’ Dimensions are Needed in Transformer
- ICML 2025Model Steering: Learning with a Reference Model Improves Generalization Bounds and Scaling Laws
- ICML 2025Nonparametric Modern Hopfield Models
- ICML 2025Occult: Optimizing Collaborative Communications across Experts for Accelerated Parallel MoE Training and Inference
- ICML 2025Olica: Efficient Structured Pruning of Large Language Models without Retraining
- ICML 2025On Efficient Estimation of Distributional Treatment Effects under Covariate-Adaptive Randomization
- ICML 2025On the Duality between Gradient Transformations and Adapters
- ICML 2025On-the-Fly Adaptive Distillation of Transformer to Dual-State Linear Attention for Long-Context LLM Serving
- ICML 2025One Leaf Reveals the Season: Occlusion-Based Contrastive Learning with Semantic-Aware Views for Efficient Visual Representation
- ICML 2025OrthoRank: Token Selection via Sink Token Orthogonality for Efficient LLM inference
- ICML 2025Over-Tokenized Transformer: Vocabulary is Generally Worth Scaling
- ICML 2025Parameter-Efficient Fine-Tuning of State Space Models
- ICML 2025Position: Enough of Scaling LLMs! Lets Focus on Downscaling
在 Claude、Cursor 里直接问这个趋势
接上 Paperlist MCP 后，写综述、挑方向或准备投稿时，不用离开对话框就能拿到这页的数字：每年各会议的论文数、增长倍数、涨得最快的子方向和代表论文，AI 会按你的问题整理。
可以直接复制这句话问 AI
用 paperlist 查高效训练在 NeurIPS 2023–2026 的论文趋势，哪一年涨得最快？
AI 会调用的工具：get_topic_stats {"topic": "efficient-training"}
常见问题
- 2023–2026 年四大会议一共有多少篇高效训练论文？
- 共 1,765 篇，按年份分别是 2023 年 346 篇、2024 年 300 篇、2025 年 516 篇、2026 年 603 篇。
- NeurIPS 2026 有多少篇高效训练论文？
- NeurIPS 2026 有 177 篇高效训练论文，上一届是 190 篇。
- ICLR 2026 有多少篇高效训练论文？
- ICLR 2026 有 148 篇高效训练论文，上一届是 126 篇。
- ICML 2026 有多少篇高效训练论文？
- ICML 2026 有 169 篇高效训练论文，上一届是 118 篇。
- CVPR 2026 有多少篇高效训练论文？
- CVPR 2026 有 109 篇高效训练论文，上一届是 82 篇。
- 哪个会议的高效训练论文最多？
- NeurIPS，2023–2026 年共 499 篇。