Skip to content
arXiv · Statistics Machine Learning· Albert Dorador·· 4 hr agoAI score28

One Permutation Is All You Need:快速、确定性的特征重要性与模型压力测试

One Permutation Is All You Need: Fast, Deterministic Feature Importance and Model Stress-Testing

AI brief

论文提出用单一最大-最小秩最优确定性置换替代 B 次随机 Monte Carlo 置换,将置换特征重要性估计的复杂度从 O(B · n · p) 降至 O(n · p),同时消除估计方差并保持或提升与真实重要性的相关性。

Source: arXiv · Statistics Machine Learning · arxiv.org