Opening black-box models used in LIBS

Opening black-box models used in LIBS

Abstrakt

The use of multivariate data-based models has become synonymous with modern LIBS analysis, be it qualitative or quantitative [1]. Two of such techniques frequently found in the LIBS literature are support vector machines (SVM) and artificial neural networks, namely convolutional neural networks (CNNs). While both techniques have undoubtedly contributed to achieving state-of-the-art classification performance in several LIBS applications, there is a common drawback associated with both methods, namely their black-box nature. In this work, we carried out the post-hoc interpretation of SVM and CNN models trained for a classification task. SVM classifiers were interpreted via the determination of feature importances [2]. The CNNs were interpreted by finding the optimal input spectra that maximize the activation of individual convolutional neurons and by carrying out class activation maximization [3]. The latter technique finds the input spectra that best represent the classes learnt by the network. We found that both classification machine learning techniques are capable of learning meaningful spectroscopic features.

The use of multivariate data-based models has become synonymous with modern LIBS analysis, be it qualitative or quantitative [1]. Two of such techniques frequently found in the LIBS literature are support vector machines (SVM) and artificial neural networks, namely convolutional neural networks (CNNs). While both techniques have undoubtedly contributed to achieving state-of-the-art classification performance in several LIBS applications, there is a common drawback associated with both methods, namely their black-box nature. In this work, we carried out the post-hoc interpretation of SVM and CNN models trained for a classification task. SVM classifiers were interpreted via the determination of feature importances [2]. The CNNs were interpreted by finding the optimal input spectra that maximize the activation of individual convolutional neurons and by carrying out class activation maximization [3]. The latter technique finds the input spectra that best represent the classes learnt by the network. We found that both classification machine learning techniques are capable of learning meaningful spectroscopic features.

machine learning, interpretability, support vector machines, spectroscopic data, convolutional neural networks

machine learning, interpretability, support vector machines, spectroscopic data, convolutional neural networks

KÉPEŠ, E.; VRÁBEL, J.; POŘÍZKA, P.; KAISER, J.

2022

24.08.2021