Is it popular or effective to concatenate the results of mean-pooling and max-pooling, to get the invariance of the latter and the expressivity of the former?