Towards uncertainty quantification in deep neural networks predicting X-ray absorption spectra Sneha Verma and Thomas Penfold Newcastle University, UK X-ray absorption near-edge structure (XANES) spectra has been widely used across the natural sciences to provide insight into the local atomic and electronic structure around an absorbing atom [1]. However quantitative interpretation can be challenging and time-consuming due to the complexity of the underlying theory. Recently, we have developed XANESNET [2,3], a deep neural network which can forecast spectral intensities only the local coordinate geometry of the transition metal complexes represented in a feature vector of weighted atom-centered symmetry functions (wACSF). However, this gives rise to a key question – how do we determine the accuracy of the predictions made from XANESNET? In the present work we extend this model to quantify the uncertainty arising from the predictions made using these models. We apply the deep ensembles and Bootstrap resampling approaches to assess the uncertainty of the X-ray absorption spectra of third row transition metal complexes. The performance of the resulting models is demonstrated by strong correlation between the predicted uncertainty and the mean square error between the actual and the predicted spectra. We also extend the XANESNET model from a multi-layer perceptron to include convolutional neural networks and autoencoders and assess the advantages and disadvantages of the performance each of these models. References 1. Guda, A. A., Guda, S. A., Martini, A., Kravtsova, A. N., Algasov, A., Bugaev, A., Kubrin, S. P., Guda, L.V., Šot, P., Van Bokhoven, J. A. and Copéret, C., 2021. Understanding X-ray absorption spectra by means of descriptors and machine learning algorithms. Npj Computational Materials,7(1), pp.1-13. 2. Penfold, T. J. and Rankine, C. D., 2022. A deep neural network for valence-to-core X-ray emission spectroscopy.Molecular Physics, p.e2123406. 3. Rankine, C. D. and Penfold, T. J., 2022. Accurate, affordable, and generalizable machine learning simulations of transition metal x-ray absorption spectra using the XANESNET deep neural network.The Journal of Chemical Physics,156(16), p.164102.
DC08
© The Author(s), 2023
Made with FlippingBook Learn more on our blog