Interpreting automatic AGN classifiers with saliency maps

Abstract

The classification of the optical spectra of active galactic nuclei (AGN) into different types is well founded on AGN physics, but it involves some degree of human oversight and cannot be reliably scaled to large data sets. Machine learning (ML) tackles such a classification problem in a fast and reproducible way, but is often perceived as a black box. However, ML interpretability and explainability are active research areas in computer science, increasingly providing us with tools to alleviate this issue. We applied ML interpretability tools to a classifier trained to predict AGN type from spectra, to demonstrate the use of such tools in this context. We trained a support-vector machine on 3346 high-quality, low redshift AGN spectra from SDSS DR15 with an existing reliable classification as type 1, type 2, or intermediate type. On a selection of test-set spectra, we computed the gradient of the predicted class probability and we built saliency maps. We also visualized the high-dimensional space of AGN spectra using t-distributed stochastic neighbor embedding (t-SNE), showing where the spectra for which we computed a saliency map are located. Regions that affect the predicted AGN type often coincide with physically relevant features, such as spectral lines. t-SNE visualization shows good separability of type 1 and type 2 spectra, while intermediate-type spectra either lie in-between as expected or appear mixed with type 2 spectra. Saliency maps show why a given AGN type was predicted by our classifier, resulting in a physical interpretation in terms of regions of the spectrum that affected its decision, making it no longer a black box. These regions coincide with those used by human experts such as relevant spectral lines, and are even used in a similar way, with the classifier e.g. effectively measuring the width of a line by weighing its center and its tails oppositely.

arXiv commentUnder review by A&A

References in corpus (26)

• Scikit-learn: Machine Learning in Python
• Unified Schemes for Radio-Loud Active Galactic Nuclei
• The Host Galaxies and Classification of Active Galactic Nuclei
• Active Galactic Nuclei: what's in a name?
• A Comprehensive Study of 2 000 Narrow Line Seyfert 1 Galaxies from the Sloan Digital Sky Survey: I. The Sample
• Eigenvector 1: An Optimal Correlation Space for Active Galactic Nuclei
• Spectral Classification of Quasars in the Sloan Digital Sky Survey: Eigenspectra; Redshift and Luminosity Effects
• CIV 1549 as an Eigenvector 1 Parameter for Active Galactic Nuclei
• Distributions of Galaxy Spectral Types in the Sloan Digital Sky Survey
• On the Unification of Active Galactic Nuclei
• Surveying the reach and maturity of machine learning and artificial intelligence in astronomy
• Dissecting stellar chemical abundance space with t-SNE
• Is Mg II 2800 a Reliable Virial Broadening Estimator for Quasars?
• Dimensionality Reduction of SDSS Spectra with Variational Autoencoders
• Spectral properties of the narrow-line region in Seyfert galaxies selected from the SDSS-DR7
• A Machine Learning Based Morphological Classification of 14,245 Radio AGNs Selected From The Best-Heckman Sample
• The GALAH survey: Chemical Tagging of Star Clusters and New Members in the Pleiades
• Removing Astrophysics in 21 cm maps with Neural Networks
• A Method to Distinguish Quiescent and Dusty Star-forming Galaxies with Machine Learning
• Do Androids Dream of Magnetic Fields? Using Neural Networks to Interpret the Turbulent Interstellar Medium
• Finding Black Holes with Black Boxes -- Using Machine Learning to Identify Globular Clusters with Black Hole Subsystems
• Active galactic nuclei synapses: X-ray versus optical classifications using artificial neural networks
• SDSS J211852.96-073227.5: The first non-local, interacting, late-type intermediate Seyfert galaxy with relativistic jets
• Applying saliency-map analysis in searches for pulsars and fast radio bursts
• Correlation of Auroral Dynamics and GNSS Scintillation with an Autoencoder
• Neural Networks and the Classification of Active Galactic Nucleus Spectra

Cited by in corpus (5)

• Explaining deep learning of galaxy morphology with saliency mapping
• The Main Sequence of quasars: the taming of the extremes
• Psi-GAN: A power-spectrum-informed generative adversarial network for the emulation of large-scale structure maps across cosmologies and redshifts
• Sparse logistic regression for RR Lyrae vs binaries classification
• Unravelling the origin of extended radio emission in narrow-line Seyfert 1 galaxies with the JVLA