平方损失下的最近邻预测理论

§1.引言 设在R~d×R~1(d≥1)取值的变量(x,θ),(x_i,θ_i),i=1,…,n相互独立,此处(X_i,θ_i)是已知样本,X之值已观测,而要依据它们去预测θ之值。引进平方损失(θ—a)~2,即用a去预测θ时,所蒙受的损失。 若知道了(x,θ)的联合分布,则风险最小的预测,即Bayes预测 δ(x)=E(θ|X=x),可无需求助于样本(X_i,θ_i),i=1,…,n而定出。当X=x时,此预测之后验风险

ON THE THEORY OF NEAREST NEIGHBOR PREDICTION UNDER SQUARE LOSS

This paper gives a rigorous treatment of the theory of NN prediction under square loss, with a minimun requirement on the regularity conditions.