Load multidimensional (numpy) array into GPU memory and slice from it - with Keras?

Question

I came across this answer, which explains how one can load a (multidimensional) numpy array as a constant into GPU memory and slice from it (each slice being a potential batch for training) with TensorFlow. I was wondering whether it was possible to do the same with Keras (with TensorFlow backend) and, if yes, how. I may have overlooked it, but I have not seen something like it in the documentation.