Is there faster way of computing loop; max(w .* x); pool; ?

Let's assume I have a matrix x \in R[#features, #data] and matrix w \in R[#units, #features]. Currently I am computing max per unit using a loop:

for k=1:numUnits
  ww = w(k, :);
  tmp = bsxfun(@times, w', x);
  r(k, :) = max(tmp, [], 1);
end

Is there any faster, no-loop equivalent version of this code?

Answer 1

[Andrei Bobrov]

r = max(bsxfun(@times,reshape(w',size(w,1),[]),reshape(x,1,size(x,2),[])),[],3);

or

r = max(bsxfun(@times,permute(w,[2 3 1]),permute(x,[3 2 1])),[],3);