偏见缓解论文趋势 2023–2026

主要结论

• 偏见缓解论文从 2023 年的 50 篇增长到 2026 年的 76 篇，三年间扩大到约 1.5 倍。
• ICLR 是发偏见缓解论文最多的会议，2023–2026 年共 85 篇，占四个会议总量的 30%。
• 交叉方向里涨得最快的是多模态大模型：同时属于两个主题的论文从 2023 年的 0 篇增至 2026 年的 15 篇。
• 在「因果与公平」的 5 个主题里，它的 2026 年论文数排第 4；同组增长最快的因果发现是 2.4 倍，偏见缓解是 1.5 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 2026Controlling for Omitted Variable Bias in Deep Neural Networks
• ICML 2026BESplit: Bias-Compensated Split Federated Learning with Evidential Aggregation
• ICLR 2026Bi-directional Bias Attribution: Debiasing Large Language Models without Modifying Prompts
• CVPR 2026BAMI Training-Free Bias Mitigation in GUI Grounding
• NeurIPS 2025Any Large Language Model Can Be a Reliable Judge: Debiasing with a Reasoning-based Bias Detector
• ICML 2025Clipped SGD Algorithms for Performative Prediction: Tight Bounds for Stochastic Bias and Remedies
• ICLR 2025ALBAR: Adversarial Learning approach to mitigate Biases in Action Recognition
• CVPR 2025Data Distributional Properties As Inductive Bias for Systematic Generalization
• NeurIPS 2024Bayesian Identification of the Hamiltonian Inductive Bias in Dynamical Systems
• ICML 2024Bias of Stochastic Gradient Descent or the Architecture: Disentangling the Effects of Overparameterization of Neural Networks
• ICLR 2024A 2-Dimensional State Space Layer for Spatial Inductive Bias
• CVPR 2024Diffeomorphic Template Registration for Atmospheric Turbulence Mitigation
• NeurIPS 2023Fast Model DeBias with Machine Unlearning
• ICML 2023FedBR: Improving Federated Learning on Heterogeneous Data via Local Learning Bias Reduction
• ICLR 2023Bias Propagation in Federated Learning
• CVPR 2023Bias Mimicking: A Simple Sampling Approach for Bias Mitigation
• NeurIPS 2026Empirical Bayes Rebiasing
• ICML 2026DISCO: Mitigating Bias in Deep Learning with Conditional Distance Correlation
• ICLR 2026BiasFreeBench: a Benchmark for Mitigating Bias in Large Language Model Responses
• CVPR 2026Bias at the End of the Score

全部偏见缓解论文

共 24 篇，第 1 / 1 页

1. ICLR 2026BA-LoRA: Bias-Alleviating Low-Rank Adaptation to Mitigate Catastrophic Inheritance in Large Language Models
2. ICLR 2026Benchmarking Bias Mitigation Toward Fairness Without Harm from Vision to LVLMs
3. ICLR 2026Bi-directional Bias Attribution: Debiasing Large Language Models without Modifying Prompts
4. ICLR 2026BiasBusters: Uncovering and Mitigating Tool Selection Bias in Large Language Models
5. ICLR 2026BiasFreeBench: a Benchmark for Mitigating Bias in Large Language Model Responses
6. ICLR 2026Directional Convergence, Benign Overfitting of Gradient Descent in leaky ReLU two-layer Neural Networks
7. ICLR 2026Discovering alternative solutions beyond the simplicity bias in recurrent neural networks
8. ICLR 2026GRADIEND: Feature Learning within Neural Networks Exemplified through Biases
9. ICLR 2026How does the optimizer implicitly bias the model merging loss landscape?
10. ICLR 2026Hyperbolic Aware Minimization: Implicit Bias for Sparsity
11. ICLR 2026Implicit Bias of Per-sample Adam on Separable Data: Departure from the Full-batch Regime
12. ICLR 2026Implicit bias produces neural scaling laws in learning curves, from perceptrons to deep networks
13. ICLR 2026Iterative Training of Physics-Informed Neural Networks with Fourier-enhanced Features
14. ICLR 2026Minor First, Major Last: A Depth-Induced Implicit Bias of Sharpness-Aware Minimization
15. ICLR 2026Never Saddle: Reparameterized Steepest Descent as Mirror Flow
16. ICLR 2026Noise Stability of Transformer Models
17. ICLR 2026OrderDP: A Theoretically Guaranteed Lossless Dynamic Data Pruning Framework
18. ICLR 2026Saddle-to-Saddle Dynamics Explains A Simplicity Bias Across Neural Network Architectures
19. ICLR 2026Saddle-To-Saddle Dynamics in Deep ReLU Networks: Low-Rank Bias in the First Saddle Escape
20. ICLR 2026The First Impression Problem: Internal Bias Triggers Overthinking in Reasoning Models
21. ICLR 2026Transformers with Endogenous In-Context Learning: Bias Characterization and Mitigation
22. ICLR 2026Trapped by simplicity: When Transformers fail to learn from noisy features
23. ICLR 2026Variational Deep Learning via Implicit Regularization
24. ICLR 2026WRING Out The Bias: A Rotation-Based Alternative To Projection Debiasing

常见问题

2023–2026 年四大会议一共有多少篇偏见缓解论文？

共 284 篇，按年份分别是 2023 年 50 篇、2024 年 73 篇、2025 年 85 篇、2026 年 76 篇。

NeurIPS 2026 有多少篇偏见缓解论文？

NeurIPS 2026 有 14 篇偏见缓解论文，上一届是 38 篇。

ICLR 2026 有多少篇偏见缓解论文？

ICLR 2026 有 24 篇偏见缓解论文，上一届是 23 篇。

ICML 2026 有多少篇偏见缓解论文？

ICML 2026 有 24 篇偏见缓解论文，上一届是 17 篇。

CVPR 2026 有多少篇偏见缓解论文？

CVPR 2026 有 14 篇偏见缓解论文，上一届是 7 篇。

哪个会议的偏见缓解论文最多？

ICLR，2023–2026 年共 85 篇。