Graph‐based calibration transfer

• Published in: Journal of chemometrics 2021-04, Vol.35 (4), p.n/a
• Main Authors: Nikzad‐Langerodi, Ramin, Sobieczky, Florian
• Format: Article
• Language: English
• Subjects: Calibration; calibration transfer; graph Laplacian; manifold regularization; partial least squares; Regularization; Standardization; transfer learning
• ISSN: 0886-9383; 1099-128X
• DOI: 10.1002/cem.3319

The problem of transferring calibrations from a primary to a secondary instrument, that is, calibration transfer (CT), has been a matter of considerable research in chemometrics over the past decades. Current state‐of‐the‐art (SoA) methods like (piecewise) direct standardization perform well when suitable transfer standards are available. However, stable calibration standards that share similar (spectral) features with the calibration samples are not always available. Towards enabling CT with arbitrary calibration standards, we propose a novel CT technique that employs manifold regularization of the partial least squares (PLS) objective. In particular, our method enforces that calibration standards, measured on primary and secondary instruments, have (nearly) invariant projections in the latent variable space of the primary calibration model. Thereby, our approach implicitly removes interdevice variation in the predictive directions of X, which is in contrast to most SoA techniques that employ explicit preprocessing of the input data. We test our approach on the well‐known corn benchmark data set employing the National Bureau of Standards and Technology (NIST) glass standard spectra for instrument standardization and compare the results with current SoA methods. We address the problem of transferring calibrations from a primary to a secondary instrument by employing manifold regularization of the partial least squares (PLS) objective. Our method enforces that calibration standards, measured on primary and secondary instruments, have (nearly) invariant projections in the latent variable space of the primary calibration model. In contrast to current state‐of‐the‐art methods like direct standardization, our approach performs well even when the transfer standards do not share similar (spectral) features with the calibration samples.