Addressing the sparsity of laser-induced breakdown spectroscopy data with randomized sparse principal component analysis

Abstract

Randomized sparse principal component analysis is more interpretable and is 20 times faster compared to regular PCA for LIBS.

Improving laser-induced breakdown spectroscopy regression modelsviatransfer learning

“…The loss function presented in (1) was used for training the transformation model as well. Owing to the sparse nature of LIBS spectra 87 every layer of the model was regularized. The same ReLU activation function was used here as well.…”

Classification of gaseous UF6 assay by femtosecond LIBS in the 424.4 nm spectral region using numerical HOGSVD-DTW features

“…The sparsity of LIBS samples relative to the number of discrete wavelength datapoints (here n = 339) is a challenge in postacquisition processing of data. 6 Since we are using higherorder singular value methods which require that data matrices have full rank, our approach to the sparsity problem is the statistical method of bootstrapping. 4,5,7,8 Bootstrapping is a technique in sampling to create additional representative samples from a given dataset in a statistically controlled way.…”

Atomic spectrometry update: review of advances in atomic spectrometry and related techniques

“…LIBS spectra are considered to exhibit high dimensionality, redundancy, and sparsity. In this context, Képeš et al 140 proposed the use of sparse PCA for the analysis of high-dimensional sparse data , significantly improving the interpretability of loading spectra. The randomised algorithm was demonstrated to offer a 20-fold increase in computation speed compared to PCA based on singular value decomposition.…”