New Boosting Methodsof GaussianProcessesfor
Yangqiu Song Regression, Changshui Zhang · 2005
Feedforward neural networks arepopular tools fornonlinear regression andclassification problems. Gaussian Process(GP) canbeviewedasanRBF neural network which haveinfinite numberofhidden neurons. Onregression problems, theycanpredict boththemeanvalueandthevariance ofthe givensample. Boosting isoneofthemostimportant recent developments inmachine learning. Classification problems have dominated research onboosting todate. Ontheother hand, the application ofboosting ofregression hasreceived less investiga- tion. Inthis paper, wedevelop twoboosting methods ofGPsfor regression according tothecharacteristic ofthem.We compare theperformance ofourensembles withother boosting algorithms andfind thatourmethods aremorestable andessentially have less over-fitting problems thantheother methods.