Has $p$-schatten pooling been done in CNN in deep neural networks?

24 Views Asked by At

In the latest 10 or so years developments in deep neural networks Convolutional Neural Networks (CNN)s have been very popular. In these networks there is a specific procedure called pooling, which this question relates to.

Given that both Max-pooling and Average-pooling seem to be used in CNNs and the fact that the $p$-schatten norms have a max-approximating behaviour as $p$ grows without bounds:

$$\lim_{p\to \infty}\|v\|_p = \lim_{p\to \infty}\left\{\left({v_1}^p+\cdots+ {v_n}^p\right)^{1/p}\right\} \to \max(v)$$

I was wondering if any kind of $p$-schatten pooling has been tried as something of a smoother variant of max-pooling.