580 California St., Suite 400
San Francisco, CA, 94104
Figure 1 – uploaded by Stéphan Clémençon
Figure 1 Incidentally, we point out that the empirical MV curve of the scoring function produced by the algorithm above is always convex, just like the target MV™ (see Proposition 4). Borrowing standard concepts of the finite element method, consider the ”hat functions”: Pe(-) = PO; (@x-1,%)) — VCs (Qe, An41)), for LS k < K, with v(a,(a’,a@”)) = (a—a’)/(a” — a’) - {a € [a’,a”"|} for a’ < a”, and set WK = (-;(ax,1)). We may then write: