Neural Network-Based Active Learning in Multivariate Calibration

Abstract

Abstract-In chemometrics, data from infrared or near-infrared (NIR) spectroscopy are often used to identify a compound or to analyze the composition of a material. This involves the calibration of models that predict the concentration of material constituents from the measured NIR spectrum. An interesting aspect of multivariate calibration is to achieve a particular accuracy level with a minimum number of training samples, as this reduces the number of laboratory tests and thus the cost of model building. In these chemometric models, the input refers to a proper representation of the spectra and the output to the concentrations of the sample constituents. The search for a most informative new calibration sample thus has to be performed in the output space of the model, rather than in the input space as in conventional modeling problems. In this paper, we propose to solve the corresponding inversion problem by utilizing the disagreements of an ensemble of neural networks to represent the prediction error in the unexplored component space. The next calibration sample is then chosen at a composition where the individual models of the ensemble disagree most. The results obtained for a realistic chemometric calibration example show that the proposed active learning can achieve a given calibration accuracy with less training samples than random sampling.Index Terms-Active learning, chemometrics, design of experiment, ensemble of models, error prediction, inverse model, model building, multivariate calibration, near infrared (NIR), spectra, spectrometer, spectroscopy.

“Inverting” X-ray Absorption Spectra of Catalysts by Machine Learning in Search for Activity Descriptors

“…Some automated procedures for selection of training examples for a particular problem may be helpful ("active learning" approach). 101,102 Before discussing NN-EXAFS and NN-XANES/EELS approaches, let us note here that ML-based regression methods can be, of course, used in the interpretation of other experimental data. For example, ML methods, trained on experimental 103−105 or simulated 106 data are used for determination of chemical shifts in NMR analysis.…”

“…Generation of training data is the most time-consuming part of NN-XAFS methods. Some automated procedures for selection of training examples for a particular problem may be helpful (“active learning” approach). , …”

Calibration model maintenance in melamine resin production: Integrating drift detection, smart sample selection and model adaptation

“…Other works on self-adaptive (evolving) chemometric models have been proposed in [10] [11], but also require the full knowledge of target measurements and are thus not applicable in our context. Active learning for off-line chemometric calibration has been proposed in [13] [14]. However, these approaches are not directly applicable for online model adaptation.…”

Low cost color assessment of turbid liquids using supervised learning data analysis – Proof of concept

“…Machine learning algorithms will be used, as a proof of concept approach, and a comparison stating the advantages and disadvantages of each algorithm is performed. Simple calibration regression models, expectation maximization Gaussian mixture (EMGM) [15][16][17] and artificial neural networks (ANN) [18,19] are used to correlate prior labeled data, this is, using supervised learning. The comparison will be performed considering features like easy of application, time of processing and error estimation by using random sampling cross validation.…”