基础模型适配论文趋势 2023–2026

主要结论

• 基础模型适配论文从 2023 年的 159 篇增长到 2026 年的 602 篇，三年间扩大到约 3.8 倍。
• NeurIPS 是发基础模型适配论文最多的会议，2023–2026 年共 531 篇，占四个会议总量的 36%。
• 交叉方向里涨得最快的是深度学习 · foundation models：同时属于两个主题的论文从 2023 年的 7 篇增至 2026 年的 66 篇。
• 在「元学习与迁移」的 5 个主题里，它的 2026 年论文数排第 1；同组增长最快的持续学习是 1.7 倍，基础模型适配是 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 2026(Mis)generalization of Helpful-Only Fine-Tuning
• ICML 2026Active Learning with Foundation Model Priors: Efficient Learning under Class Imbalance
• ICLR 2026Benchmarking ECG Foundational Models: A Reality Check Across Clinical Tasks
• CVPR 2026CD-Buffer Complementary Dual-Buffer Framework for Test-Time Adaptation in Adverse Weather
• NeurIPS 2025An Investigation of Memorization Risk in Healthcare Foundation Models
• ICML 2025A Bayesian Model Selection Criterion for Selecting Pretraining Checkpoints
• ICLR 2025Better autoregressive regression with LLMs via regression-aware fine-tuning
• CVPR 20255%>100%: Breaking Performance Shackles of Full Fine-Tuning on Visual Recognition Tasks
• NeurIPS 2024Instruction Tuning Large Language Models to Understand Electronic Health Records
• ICML 2024By Tying Embeddings You Are Assuming the Distributional Hypothesis
• ICLR 2024A Hard-to-Beat Baseline for Training-free CLIP-based Adaptation
• CVPR 2024DePT: Decoupled Prompt Tuning
• NeurIPS 2023A Fast and Accurate Estimator for Large Scale Linear Model via Data Averaging
• ICML 2023Are Neurons Actually Collapsed? On the Fine-Grained Structure in Neural Representations
• ICLR 2023A Closer Look at Model Adaptation using Feature Distortion and Simplicity Bias
• CVPR 2023A-La-Carte Prompt Tuning (APT): Combining Distinct Data via Composable Prompting
• NeurIPS 2026A Graph Foundation Model for Unified Clustering
• ICML 2026AdaMeZO: Adam-style Zeroth-Order Optimizer for LLM Fine-tuning Without Maintaining the Moments
• ICLR 2026Bridging Radiology and Pathology Foundation Models via Concept-Based Multimodal Co-Adaptation
• CVPR 2026Fed-ADE Adaptive Learning Rate for Federated Post-adaptation under Distribution Shift

全部基础模型适配论文

共 324 篇，第 3 / 7 页

101. ICML 2026Rethinking the Flow-based Gradual Domain Adaptation: A Semi-Dual Optimal Transport Perspective
102. ICML 2026SALAAD: Sparse And Low-Rank Adaptation via ADMM for Large Language Model Inference
103. ICML 2026Saving Foundation Flow-Matching Priors for Inverse Problems
104. ICML 2026Scaling Behavior in Model Fine-tuning for Audio DeepFake Detection
105. ICML 2026ScaLoRA: Optimally Scaled Low-Rank Adaptation for Efficient High-Rank Fine-Tuning
106. ICML 2026scDEBART: Predicting in silico Single-Cell Perturbation Responses via Large-Scale Differential Expression Learning
107. ICML 2026See More, Forecast Better and Faster: Enhancing Time Series Foundation Models via Inference-Time Plug-and-Play Downsampling
108. ICML 2026SHERPA: Fine-tuning Segment Anything Models with Task-relevant Guidance
109. ICML 2026SHINE: A Scalable In-Context Hypernetwork for Mapping Context to LoRA in a Single Pass
110. ICML 2026SleepMaMi: A Universal Sleep Foundation Model for Integrating Macro- and Micro-structures
111. ICML 2026SliceFine: The Universal Winning-Slice Hypothesis for Pretrained Networks
112. ICML 2026SpanNorm: Reconciling Training Stability and Performance in Deep Transformers
113. ICML 2026SPARD: Defending Harmful Fine-Tuning Attack via Safety Projection with Relevance–Diversity Data Selection
114. ICML 2026Spectral Imbalance Causes Forgetting in Low-Rank Continual Adaptation
115. ICML 2026SPR: A Structured Prompt Refinement Network for Modality Missing
116. ICML 2026StarEmbed: Benchmarking Time Series Foundation Models on Astronomical Observations of Variable Stars
117. ICML 2026Subgroup Discovery with the Cox Model
118. ICML 2026SURGE: Surrogate Gradient Adaptation in Binary Neural Networks
119. ICML 2026Surgery: Mitigating Harmful Fine-Tuning for Large Language Models via Attention Sink
120. ICML 2026TabICLv2: A Better, Faster, Scalable, and Open Tabular Foundation Model
121. ICML 2026TabMGP: Martingale Posterior with TabPFN
122. ICML 2026TabSwift: An Efficient Tabular Foundation Model with Row-Wise Attention
123. ICML 2026TarGATE: Target-Aware Data Selection via Token-Attenuation Gates
124. ICML 2026The Geometry of Narrow Fine-Tuning Degradation: Trajectory Lock-in and Spectral Bifurcation
125. ICML 2026The Geometry of Sequential Learning: Lie-Bracket Prediction of Transfer Order
126. ICML 2026Time-PEFT: Temporal and Multichannel Complexity-Based Fine-Tuning for Time-Series Foundation Models
127. ICML 2026Towards Fine-Grained Robustness: Attention-Guided Test-Time Prompt Tuning for Vision-Language Models
128. ICML 2026Towards Understanding the Dynamics of Low-Rank Adaptation
129. ICML 2026Trust It or Not: Evidential Uncertainty for Feed-Forward 3D Reconstruction with Trust3R
130. ICML 2026TuneAhead: Predicting Fine-tuning Performance Before Training Begins
131. ICML 2026UB-SMoE: Universally Balanced Sparse Mixture-of-Experts for Resource-adaptive Federated Fine-tuning of Foundation Models
132. ICML 2026Uncovering the Latent Potential of Deep Intermediate Representations
133. ICML 2026Universal Redundancies in Time Series Foundation Models
134. ICML 2026Unveiling Prior-Data Fitted Networks on Causal Effect Estimation: Pre-Training or Fine-Tuning?
135. ICML 2026Utility-Diversity Aware Online Batch Selection for LLM Supervised Fine-tuning
136. ICML 2026Walrus: A Cross-domain Foundation Model for Continuum Dynamics
137. ICML 2026Weight Updates as Activation Shifts: A Principled Framework for Steering
138. ICML 2026When Do Graph Foundation Models Transfer? A Data-Centric Theory
139. ICML 2026When More Data Doesn't Help: Limits of Adaptation in Multitask Learning
140. ICML 2026When Tabular Foundation Models Meet Strategic Tabular Data: A Prior Alignment Approach
141. ICML 2026WorldMirror: Universal 3D World Reconstruction with Any-Prior Prompting
142. ICML 2026Zeus: Towards Tuning-Free Foundation Model for Time Series Analysis
143. ICML 2025A Bayesian Model Selection Criterion for Selecting Pretraining Checkpoints
144. ICML 2025A Stronger Mixture of Low-Rank Experts for Fine-Tuning Foundation Models
145. ICML 2025Adapting to Evolving Adversaries with Regularized Continual Robust Training
146. ICML 2025AdaPTS: Adapting Univariate Foundation Models to Probabilistic Multivariate Time Series Forecasting
147. ICML 2025Are Large Brainwave Foundation Models Capable Yet ? Insights from Fine-Tuning
148. ICML 2025Benign Samples Matter! Fine-tuning On Outlier Benign Samples Severely Breaks Safety
149. ICML 2025Beyond Sensor Data: Foundation Models of Behavioral Data from Wearables Improve Health Predictions
150. ICML 2025Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics

常见问题

2023–2026 年四大会议一共有多少篇基础模型适配论文？

共 1,492 篇，按年份分别是 2023 年 159 篇、2024 年 270 篇、2025 年 461 篇、2026 年 602 篇。

NeurIPS 2026 有多少篇基础模型适配论文？

NeurIPS 2026 有 182 篇基础模型适配论文，上一届是 174 篇。

ICLR 2026 有多少篇基础模型适配论文？

ICLR 2026 有 187 篇基础模型适配论文，上一届是 149 篇。

ICML 2026 有多少篇基础模型适配论文？

ICML 2026 有 142 篇基础模型适配论文，上一届是 76 篇。

CVPR 2026 有多少篇基础模型适配论文？

CVPR 2026 有 91 篇基础模型适配论文，上一届是 62 篇。

哪个会议的基础模型适配论文最多？

NeurIPS，2023–2026 年共 531 篇。