a b s t r a c t
Extre me learning m achine (ELM) is an effic ient learn ing algorith m for genera lized single hidden lay er
feedf orwar d netw orks (SLF Ns), wh ich perf orms we ll in both re gressio n and classi ficati on app lications .
It has recently been show n that from the optimi zation point of view ELM and supp ort vector machi ne
(SVM ) are equivale nt but ELM has less strin gent optimi zation constra ints. Due to the mild optimi zation
const raints ELM can be easy of impleme ntatio n and usu ally ob tains better ge neraliza tion perf orma nce.
In this pap er we study the perf orma nce of the one-aga inst- all (OAA ) and one-aga inst- one (O AO) ELM
for classi ficati on in m ulti-lab el face re cognit ion applica tions. The perfo rmance is verifie d through four
benc hmarki ng face image data sets.
& 2011 Elsev ier B.V. All rights re served.