Sure. Assuming all your training vectors are in rows of a 2D array, did you try something like this
for k = 1 : n
% Extract the k'th training vector from the list of all of them.
thisTrainingVector = allTrainingVectors(k, :); % Extract row
% Get difference between this vector and the first "reconstruction vector".
diff1 = thisTrainingVector - reconstructionVector1;
% Get difference between this vector and the second "reconstruction vector".
diff2 = thisTrainingVector - reconstructionVector2;
% Then do whatever you need to do with diff1 and diff2.
end