1: \begin{abstract}

2: We consider the functional linear regression model with a scalar response and a Hilbert space-valued predictor, a well-known ill-posed inverse problem. We propose a new formulation of the functional partial least-squares (PLS) estimator related to the conjugate gradient method. We shall show that the estimator achieves the (nearly) optimal convergence rate on a class of ellipsoids and we introduce an early stopping rule which adapts to the unknown degree of ill-posedness. Some theoretical and simulation comparison between the estimator and the principal component regression estimator is provided.

3: 

4: \bigskip

5: 

6: \textbf{Keywords:} Adaptive Estimation, Conjugate Gradient Method, Convergence Rates, Functional Linear Regression, Functional Partial Least Squares.\\

7: \textbf{JEL classification:} C38, C51, C53, C55, C58.

8: 

9: \end{abstract}

10: