AUTOMATED ASSIGNMENT OF ROTATIONAL SPECTRA USING ARTIFICIAL NEURAL NETWORKS

Zaleski, Daniel P.

Permalink

https://hdl.handle.net/2142/100653 Copy

Description

Title

AUTOMATED ASSIGNMENT OF ROTATIONAL SPECTRA USING ARTIFICIAL NEURAL NETWORKS

Author(s)

Zaleski, Daniel P.

Contributor(s)

Prozument, Kirill

Issue Date

06/19/18

Keyword(s)

Mini-symposium: New Ways of Understanding Molecular Spectra

Abstract

Last year at this conference several approaches to utilize machine learning\footnote{Bishop, C M. “Neural networks for pattern recognition.” Oxford university press, 1995.} to train a computer to recognize the patterns inherit in rotational spectra were presented\footnote{Zaleski, D. P.; Prozument, K. Identifying Broadband Rotational Spectra with Neural Networks, International Symposium on Molecular Spectroscopy, June 21; 2017.}. It was shown that the recognized patterns could be used to identify (or classify) a rotational spectrum by its Hamiltonian type, but at the time, the rotational constants were not recovered. Here, we describe a feed forward artificial neural network that has been trained to identify different types of rotational spectra and determine the parameters of the molecular Hamiltonians. The network requires no user interaction beyond loading a “peak pick”, and can return fits within a fraction of a second. The rotational constants are typically deduced with the accuracy of 1–10 MHz. We will describe how the network works and provide benchmarking results.

Publisher

International Symposium on Molecular Spectroscopy

Type of Resource

text

Language

English

Permalink

http://hdl.handle.net/2142/100653

DOI

10.15278/isms.2018.TH01

Owning Collections