-
Stellar Populations With Optical Spectra: Deep Learning vs. Popular Spectrum Fitting Codes
Authors:
Joanna Woo,
Dan Walters,
Finn Archinuk,
S. M. Faber,
Sara L. Ellison,
Hossen Teimoorinia,
Kartheik Iyer
Abstract:
We compare the performance of several popular spectrum fitting codes (Firefly, starlight, pyPipe3D and pPXF), and a deep-learning convolutional neural network (StarNet), in recovering known stellar population properties (mean stellar age, stellar metallicity, stellar mass-to-light ratio M*/L_r and the internal E(B-V)) of simulated galaxy spectra in optical wavelengths. Our mock spectra are constru…
▽ More
We compare the performance of several popular spectrum fitting codes (Firefly, starlight, pyPipe3D and pPXF), and a deep-learning convolutional neural network (StarNet), in recovering known stellar population properties (mean stellar age, stellar metallicity, stellar mass-to-light ratio M*/L_r and the internal E(B-V)) of simulated galaxy spectra in optical wavelengths. Our mock spectra are constructed from star-formation histories from the IllustrisTNG100-1 simulation. These spectra mimic the Sloan Digital Sky Survey (SDSS) through a novel method of including the noise, sky residuals and emission lines taken directly from SDSS. We find that StarNet vastly outperforms all conventional codes in both speed and recovery of stellar population properties (error scatter < 0.08 dex, average biases < 0.02 dex for all tested quantities), but it requires an appropriate training set. Of the non-machine-learning codes, pPXF was a factor of 3-4 times faster than the other codes, and was the best in recovering stellar population properties (error scatter of < 0.11 dex, average biases < 0.08 dex). However, the errors and biases are strongly dependent on both true and predicted values of stellar age and metallicity, and signal-to-noise ratio. The biases of all codes can approach 0.15 dex in stellar ages, metallicities and log M*/L_r , but remain < 0.05 for E(B-V). Using unrealistic Gaussian noise in the construction of mock spectra will underestimate the errors in the metallicities by a factor of two or more, and mocks without emission lines will underestimate the errors in stellar age and M*/L_r by a factor of two.
△ Less
Submitted 25 April, 2024; v1 submitted 22 January, 2024;
originally announced January 2024.
-
Mitigating the Non-Linearities in a Pyramid Wavefront Sensor
Authors:
Finn Archinuk,
Rehan Hafeez,
Sébastien Fabbro,
Hossen Teimoorinia,
Jean-Pierre Véran
Abstract:
For natural guide start adaptive optics (AO) systems, pyramid wavefront sensors (PWFSs) can provide significant increase in sensitivity over the traditional Shack-Hartmann, but at the cost of a reduced linear range. When using a linear reconstructor, non-linearities result in wavefront estimation errors, which can have a significant impact on the image quality delivered by the AO system. Here we s…
▽ More
For natural guide start adaptive optics (AO) systems, pyramid wavefront sensors (PWFSs) can provide significant increase in sensitivity over the traditional Shack-Hartmann, but at the cost of a reduced linear range. When using a linear reconstructor, non-linearities result in wavefront estimation errors, which can have a significant impact on the image quality delivered by the AO system. Here we simulate a wavefront passing through a PWFS under varying observing conditions to explore the possibility of using a non-linear machine learning model to estimate wavefront errors better than a linear reconstruction. We find significant improvement even with light-weight models, underscoring the need for further investigation of this approach.
△ Less
Submitted 1 December, 2023; v1 submitted 16 May, 2023;
originally announced May 2023.
-
Mapping the Diversity of Galaxy Spectra with Deep Unsupervised Machine Learning
Authors:
Hossen Teimoorinia,
Finn Archinuk,
Joanna Woo,
Sara Shishehchi,
Asa F. L. Bluck
Abstract:
Modern spectroscopic surveys of galaxies such as MaNGA consist of millions of diverse spectra covering different regions of thousands of galaxies. We propose and implement a deep unsupervised machine learning method to summarize the entire diversity of MaNGA spectra onto a 15x15 map (DESOM-1), where neighbouring points on the map represent similar spectra. We demonstrate our method as an alternati…
▽ More
Modern spectroscopic surveys of galaxies such as MaNGA consist of millions of diverse spectra covering different regions of thousands of galaxies. We propose and implement a deep unsupervised machine learning method to summarize the entire diversity of MaNGA spectra onto a 15x15 map (DESOM-1), where neighbouring points on the map represent similar spectra. We demonstrate our method as an alternative to conventional full spectral fitting for deriving physical quantities, as well as their full probability distributions, much more efficiently than traditional resource-intensive Bayesian methods. Since spectra are grouped by similarity, the distribution of spectra onto the map for a single galaxy, i.e., its "fingerprint", reveals the presence of distinct stellar populations within the galaxy indicating smoother or episodic star-formation histories. We further map the diversity of galaxy fingerprints onto a second map (DESOM-2). Using galaxy images and independent measures of galaxy morphology, we confirm that galaxies with similar fingerprints have similar morphologies and inclination angles. Since morphological information was not used in the mapping algorithm, relating galaxy morphology to the star-formation histories encoded in the fingerprints is one example of how the DESOM maps can be used to make scientific inferences.
△ Less
Submitted 6 December, 2021;
originally announced December 2021.
-
Forecasting Wavefront Corrections in an Adaptive Optics System
Authors:
Rehan Hafeez,
Finn Archinuk,
Sébastien Fabbro,
Hossen Teimoorinia,
Jean-Pierre Véran
Abstract:
We use telemetry data from the Gemini North ALTAIR adaptive optics system to investigate how well the commands for wavefront correction (both Tip/Tilt and high-order turbulence) can be forecasted in order to reduce lag error (due to wavefront sensor averaging and computational delays) and improve delivered image quality. We show that a high level of reduction ($\sim$ 5 for Tip-Tilt and $\sim$ 2 fo…
▽ More
We use telemetry data from the Gemini North ALTAIR adaptive optics system to investigate how well the commands for wavefront correction (both Tip/Tilt and high-order turbulence) can be forecasted in order to reduce lag error (due to wavefront sensor averaging and computational delays) and improve delivered image quality. We show that a high level of reduction ($\sim$ 5 for Tip-Tilt and $\sim$ 2 for high-order modes) in RMS wavefront error can be achieved by using a "forecasting filter" based on a linear auto-regressive model with only a few coefficients ($\sim$ 30 for Tip-Tilt and $\sim$ 5 for high-order modes) to complement the existing integral servo-controller. Updating this filter to adapt to evolving observing conditions is computationally inexpensive and requires less than 10 seconds worth of telemetry data. We also use several machine learning models (Long-Short Term Memory and dilated convolutional models) to evaluate whether further improvements could be achieved with a more sophisticated non-linear model. Our attempts showed no perceptible improvements over linear auto-regressive predictions, even for large lags where residuals from the linear models are high, suggesting that non-linear wavefront distortions for ALTAIR at the Gemini North telescope may not be forecasted with the current setup
△ Less
Submitted 12 May, 2022; v1 submitted 2 December, 2021;
originally announced December 2021.
-
An astronomical image content-based recommendation system using combined deep learning models in a fully unsupervised mode
Authors:
Hossen Teimoorinia,
Sara Shishehchi,
Ahnaf Tazwar,
Ping Lin,
Finn Archinuk,
Stephen D. J. Gwyn,
J. J. Kavelaars
Abstract:
We have developed a method that maps large astronomical images onto a two-dimensional map and clusters them. A combination of various state-of-the-art machine learning (ML) algorithms is used to develop a fully unsupervised image quality assessment and clustering system. Our pipeline consists of a data pre-processing step where individual image objects are identified in a large astronomical image…
▽ More
We have developed a method that maps large astronomical images onto a two-dimensional map and clusters them. A combination of various state-of-the-art machine learning (ML) algorithms is used to develop a fully unsupervised image quality assessment and clustering system. Our pipeline consists of a data pre-processing step where individual image objects are identified in a large astronomical image and converted to smaller pixel images. This data is then fed to a deep convolutional autoencoder jointly trained with a self-organizing map (SOM). This part can be used as a recommendation system. The resulting output is eventually mapped onto a two-dimensional grid using a second, deep, SOM. We use data taken from ground-based telescopes and, as a case study, compare the system's ability and performance with the results obtained by supervised methods presented by Teimoorinia et al. (2020). The availability of target labels in this data allowed a comprehensive performance comparison between our unsupervised and supervised methods. In addition to image-quality assessments performed in this project, our method can have various other applications. For example, it can help experts label images in a considerably shorter time with minimum human intervention. It can also be used as a content-based recommendation system capable of filtering images based on the desired content.
△ Less
Submitted 27 February, 2021;
originally announced March 2021.
