BriefGPT.xyz
May, 2019
核方法的渐近学习曲线：实证数据 vs 教师-学生范式
Asymptotic learning curves of kernel methods: empirical data v.s. Teacher-Student paradigm
HTML
PDF
Stefano Spigler, Mario Geiger, Matthieu Wyart
TL;DR
研究使用核方法学习监督任务所需的训练数据量，发现泛化误差随训练数据量$n$的负指数幂下降，其指数$eta$依赖于数据平滑度和维度，在真实数据集上实验表明$eta$一般较小，可以通过关注真实函数在核的特征向量上的投影来预测$eta$。
Abstract
How many
training data
are needed to learn a supervised task? It is often observed that the
generalization error
decreases as $n^{-\beta}$ where $n$ is the number of training examples and $\beta$ an exponent that
→