Supplementary Materials for ” Fast Matrix - vector Multiplications for Large - scale Logistic Regression on Shared - memory Systems ”

For data sets with many features, ∇f(w) is a huge and dense matrix that is too large to be stored. Machine learning Algorithm 1 CG for approximately solving (A.4) 1) Let s = 0, r = −∇f(w), d = r, and rsq = ‖r‖. 2) For i = 0, 1, . . . • If some stopping conditions hold, stop and output s as s. • Calculate v = ∇f(w)d by (A.5) • α← rsq/dv • s← s+ αd and r ← r − αv. • rnew sq ← ‖r‖ • β ← rnew sq /rsq and rsq ← rnew sq • d← r + βd.