Abstract : The problem of adaptive estimation of the regression function f from noisy observations is concerned. A confidence interval for the L_2-error for wavelet adaptive estimator is provided. We show that if f belongs to a Besov class, the proposed confidence interval is minimax.
