Knowledge Discovery 1

Time: Thursday Apr 28 22:30 UTC

Raffaele D'Abrusco



Rafael Martinez Galarza

X-ray datasets: A Machine Learning Perspective 12'+3'

As the volume of astronomical X-ray datasets increases with the advent of dedicated catalogs and new facilities, data science and machine learning methods are becoming a useful tool for discovery in high energy astrophysics, in particular in the context of time-domain anomalies. Due to the specific nature of X-ray datasets, which are lists of individual photon detections called event files, the pre-processing practices of the data for the purpose of classification, regression, or anomaly detection need to be carefully reviewed before any existing methods designed for optical or infrared datasets is used. Here I provide an overview of our experience pre-processing X-ray datasets for machine learning applications, and adapting methods of classification and anomaly detections for X-ray catalogs (in particular, the Chandra Source Catalog, CSC). I will discuss what CSC interfaces are most useful for the seamless use of machine learning for discovery, and provide recommendations as to what other data providers should do to make these methods easier to use. Finally, I will highlight some results of our investigations.

Petr Skoda

The Role of VO Technology in Astronomical Machine

12'+3' The VO infrastructure has already increased the efficiency of traditional astronomical data analysis in many ways. We try to identify the points where the IVOA standards (current or after some modifications) can make easier the preparation of machine learning experiments or even introduce new KDD methodology.
Ying Wu

Classification of Galaxy Spectra based on Convolutional Neural Network

12'+3' It is important to classify galaxies by their spectral data in astronomy. The widely used spectral classification method for galaxy is the BPT diagram, which classifies emission line galaxies by flux ratios of the Balmer and forbidden lines. A convolutional neural network (CNN) is a deep learning algorithm which achieves outstanding performance on feature extraction and classification tasks. We build a one-dimensional CNN model to classify galaxy spectra into star-forming, composite, AGN, and normal galaxies. We preprocessed the galaxy spectra selected from SDSS DR16 with spectral interval constraint and flux standardization and conducted experiments with the preprocessed data. The dataset labels are derived from the intersection of the BPT classification given by MPA-JHU and Portsmouth catalog. The results showed that the classification accuracy of our network was over 92% without relying on redshifts.

Moderator: Raffaele D'Abrusco, Notetaker: TBD, Etherpad link

Topic attachments
I Attachment History Action Size Date Who Comment
Unknown file formatpptx IVOA2022InterOp_KDIG_Talk_Ying_Wu_220428.pptx r1 manage 1839.9 K 2022-04-28 - 11:25 YihanTao  
PDFpdf ivoa_interop_2022_martinez.pdf r1 manage 15108.5 K 2022-04-28 - 21:38 RaffaeleDAbrusco  
PDFpdf skoda-MLVO.pdf r1 manage 2151.8 K 2022-04-28 - 21:51 PetrSkoda  
Edit | Attach | Watch | Print version | History: r10 < r9 < r8 < r7 < r6 | Backlinks | Raw View | Raw edit | More topic actions
Topic revision: r10 - 2022-04-28 - PetrSkoda
This site is powered by the TWiki collaboration platform Powered by Perl This site is powered by the TWiki collaboration platformCopyright © 2008-2022 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback