There is probably a significantly faster way to implement translation.
I initially implemented translation in a dumb way, where the inverse of the single input argument y is (or might be) computed many times. (I say "might be" because I'm not sure how smart the Python interpreter is.)
Since a translation takes place for each element in the group at the map stage of convolution, a significantly faster translation will result in a significantly faster convolution.
There is probably a significantly faster way to implement translation.
I initially implemented translation in a dumb way, where the inverse of the single input argument y is (or might be) computed many times. (I say "might be" because I'm not sure how smart the Python interpreter is.)
Since a translation takes place for each element in the group at the map stage of convolution, a significantly faster translation will result in a significantly faster convolution.