We study the asymptotic overfitting behavior of interpolation with minimum norm ($\ell_2$ of the weights) two-layer ReLU networks for noisy univariate regression. We show that overfitting is tempered for the $L_1$ loss, and any $L_p$ loss for $p<2$, but catastrophic for $p\geq 2$.

翻译：我们研究了最小范数（权重的$\ell_2$范数）两层ReLU网络在带噪单变量回归中插值行为的渐近过拟合特性。结果表明：对于$L_1$损失以及所有满足$p<2$的$L_p$损失，过拟合现象受到抑制；但当$p\geq 2$时，过拟合则呈现灾难性特征。