Search papers, labs, and topics across Lattice.
This work addresses scalar-on-function linear regression under realistic measurement regimes where functional inputs are observed on discrete regular grids corrupted by additive noise. To overcome the coupled challenges of curve de-noising and ill-posed inverse regression, the authors deploy a two-step procedure combining Fourier projection reconstruction with penalized least-squares estimation over adaptive trigonometric bases. The resulting estimator satisfies non-asymptotic oracle inequalities and is proven to achieve the minimax-optimal prediction error rate under polynomial covariate eigenvalue decay when grid resolution is sufficiently high.
Functional linear regression can achieve minimax-optimal prediction rates even when covariates are masked by discrete, noisy sampling, bridging the long-standing gap between idealized continuous theory and imperfect real-world sensor streams.
In this paper, we consider the scalar-on-function linear regression model under a realistic sampling scheme in which the functional covariates are observed on a regular grid and contaminated by additive noise. We propose a two-step estimation procedure: first, the underlying curves are reconstructed from the discrete noisy observations using a Fourier-based projection method; second, the slope function is estimated by a penalized least-squares criterion over finite-dimensional trigonometric spaces, with data-driven selection of the model dimension. We establish oracle-type inequalities for the prediction error, both with respect to the reconstructed curves and to the true latent curves. Under regularity assumptions on the slope function and polynomial decay of the eigenvalues of the covariate, we derive convergence rates for the prediction error and show that our estimator attains the minimax rate when the number of grid points is sufficiently large. Finally, the proposed method is illustrated on simulated data and on a real meteorological dataset.
