I have a loss function implemented in TensorFlow that computes mean squared error. All tensors being used to compute the objective are of type float64 and therefore the loss function itself is of d...