J/A+A/708/A215 Stellar age determination using deep NN (Boin+, 2026)
Stellar age determination using deep neural networks.
Isochrone ages for 1.3 million stars, based on BaSTI, MIST, PARSEC, Dartmouth,
and SYCLIST evolutionary grids.
Boin T., Casamiquela L., Haywood M., Di Matteo P., Lebreton Y., Uddin M.,
Reese D.
<Astron. Astrophys. 708, A215 (2026)>
=2026A&A...708A.215B 2026A&A...708A.215B (SIMBAD/NED BibCode)
ADC_Keywords: Milky Way ; Photometry ; Abundances ; Isochrones ;
Models, evolutionary ; Stars, ages
Keywords: stars: fundamental parameters - Galaxy: evolution -
Galaxy: kinematics and dynamics - Galaxy: stellar content
Abstract:
Recent spectroscopic surveys provide element abundances for large
samples of Milky Way stars, from which stellar parameters can be
inferred. Stellar ages, among them, are both a notoriously difficult
parameter to estimate and a fundamental property for Galactic
archaeology studies.
We aim to develop a model-driven deep learning approach to age
determination, by training neural networks on stellar evolutionary
grids. Contrary to the usual data-driven deep learning approach of
using prior age estimates as training data, our method has the
potential for a wider and less biased range of application. The low
computational cost of deep learning methods compared to e.g., bayesian
isochrone-fitting allows for a broad analysis of large spectroscopic
catalogues.
We train multilayer perceptrons on different stellar evolutionary
grids to map [M/H], MG, (GBP-GRP) to stellar age tau. We combine
Gaia photometry and parallaxes, metallicities and alpha elements from
spectroscopic surveys and extinction maps, which are passed through
the neural networks to estimate stellar ages.
We apply our method to the LAMOST DR10, GALAH DR3 & DR4 and APOGEE
DR17 spectroscopic surveys, for which we estimate the ages using the
BaSTI tracks, along with other stellar evolutionary models. We
leverage this novel technique to study, for the first time,
differences in age estimates from several evolutionary grids applied
on very large datasets. In addition, we date 13 open clusters and one
globular cluster and find a median absolute deviation with literature
ages of 0.20Gyr. Along with the stellar ages catalogues from our
estimates, we release NEST (Neural Estimator of Stellar Times), a
python package to estimate stellar age based on this work and
available at https://github.com/star-age/NEST, as well as a web
interface https://star-age.github.io.
We show that, when using the same evolutionary grid, our method
retrieves the same ages as a bayesian approach like SPInS, for only a
fraction of the computational cost, with a 60000 speedup factor for a
typical star. This model-driven deep learning technique thus opens up
the way for broad galactic archeology studies on the largest datasets
available today and in the near future with upcoming surveys such as
4MOST.
Description:
The catalogues contain age estimations for 1.3 million stars, derived
using NEST from Gaia magnitudes and abundances, for several large
spectroscopic surveys (LAMOST DR10, GALAH DR3, GALAH DR4 and APOGEE
DR17).
File Summary:
--------------------------------------------------------------------------------
FileName Lrecl Records Explanations
--------------------------------------------------------------------------------
ReadMe 80 . This file
lamostl.dat 282 935727 LAMOST low-resolution DR10 stars age catalogue
lamostm.dat 282 88357 LAMOST medium-resolution DR10 stars age catalogue
galah3.dat 282 13441 GALAH DR3 stars age catalogue
galah4.dat 282 251659 GALAH DR4 stars ages catalogue
apogee.dat 282 147581 APOGEE DR17 stars age catalogue
--------------------------------------------------------------------------------
See also:
I/355 : Gaia DR3 Part 1. Main source (Gaia Collaboration, 2022)
V/162 : LAMOST DR11 catalogs (Luo+, 2026)
J/MNRAS/506/150 : The GALAH+ Survey DR3 (Buder+, 2021)
Byte-by-byte Description of file: lamost?.dat galah?.dat apogee.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 19 I19 --- GaiaDR3 Gaia DR3 source ID (source_id)
21- 24 F4.2 mag ABP Gaia BP-band extinction (ABP)
26- 29 F4.2 mag ARP Gaia RP-band extinction (ARP)
31- 34 F4.2 mag AG Gaia G-band extinction (AG)
36- 39 F4.2 mag Av Visual extinction (Av)
41- 46 F6.3 mag GBP-GRP0 GBP-GRP color extinction corrected
(GBPGRP0)
48- 53 F6.3 mag GMAG Absolute G-band magnitude (MG)
55- 59 F5.2 [-] [alpha/Fe] Abundance [alpha/Fe] (alpha_fe)
61- 65 F5.2 [-] [Fe/H] Abundance [Fe/H] (feh)
67- 71 F5.2 [-] [Fe/H]caled Scaled Fe/H using Salaris et al.
93 relation (feh_scaled)
73- 77 F5.2 [-] logg Surface gravity (logg)
79- 82 F4.2 [-] e_[Fe/H] Uncertainty on [Fe/H] (err_feh)
84- 87 F4.2 [-] e_[Fe/H]caled Uncertainty on scaled [Fe/H]
(errfehscaled)
89- 92 F4.2 mag e_ABP Error on ABP (eABP)
94- 97 F4.2 mag e_ARP Error on ARP (eARP)
99-102 F4.2 mag e_AG Error on AG (eAG)
104-108 F5.3 mag e_GMAG Error on MG (eMG)
110-114 F5.3 mag e_GBP-GRP0 Error on GBPGRP0 (eBPRP)
116-119 F4.2 [-] e_[alpha/Fe] Error on [alpha/Fe] (erralphafe)
121-126 F6.2 [-] e_logg ?=-99 Error on logg (logg_err)
128-134 F7.3 Gyr AgeBaSTImean Mean age using a BaSTI-trained Neural
Network (AgeBaSTImean)
136-142 F7.3 Gyr AgeBaSTImedian Median age using a BaSTI-trained
Neural Network (AgeBaSTImedian)
144-148 F5.3 Gyr AgeBaSTIstd Age standard deviation using a
BaSTI-trained Neural Network
(AgeBaSTIstd)
150-155 F6.3 Gyr AgeBaSTImode Mode age using a BaSTI-trained Neural
Network (AgeBaSTImode)
157 I1 --- inDomainBaSTI Boolean flag: is within BaSTI grid
domain (1=yes,0=no)
(indomainBaSTI)
159-164 F6.3 Gyr AgeMISTmean Mean age using a MIST-trained Neural
Network (AgeMISTmean)
166-171 F6.3 Gyr AgeMISTmedian Median age using a MIST-trained
Neural Network (AgeMISTmedian)
173-178 F6.3 Gyr AgeMISTstd Age standard deviation using a
MIST-trained Neural Network
(AgeMISTstd)
180-185 F6.3 Gyr AgeMISTmode Mode age using a MIST-trained Neural
Network (AgeMISTmode)
187 I1 --- inDomainMIST Boolean flag: is within MIST grid
domain (1=yes,0=no) (indomainMIST)
189-195 F7.3 Gyr AgePARSECmean Mean age using a PARSEC-trained
Neural Network (AgePARSECmean)
197-203 F7.3 Gyr AgePARSECmedian Median age using a PARSEC-trained
Neural Network (AgePARSECmedian)
205-210 F6.3 Gyr AgePARSECstd Age standard deviation using a
PARSEC-trained Neural Network
(AgePARSECstd)
212-217 F6.3 Gyr AgePARSECmode Mode age using a PARSEC-trained
Neural Network (AgePARSECmode)
219 I1 --- inDomainPARSEC Boolean flag: is within PARSEC grid
domain (1=yes,0=no)
(indomainPARSEC)
221-227 F7.3 Gyr AgeDartmouth_mean Mean age using a Dartmouth-trained
Neural Network (AgeDartmouthmean)
229-235 F7.3 Gyr AgeDartmouthmedian Median age using a Dartmouth-trained
Neural Network
(AgeDartmouthmedian)
237-242 F6.3 Gyr AgeDartmouthstd Age standard deviation using a
Dartmouth-trained Neural Network
(AgeDartmouthstd)
244-249 F6.3 Gyr AgeDartmouthmode Mode age using a Dartmouth-trained
Neural Network (AgeDartmouthmode)
251 I1 --- inDomainDartmouth Boolean flag: is within Dartmouth
grid domain (1=yes,0=no)
(indomainDartmouth)
253-259 F7.3 Gyr AgeGenevamean Mean age using a Geneva-trained
Neural Network (AgeGenevamean)
261-267 F7.3 Gyr AgeGenevamedian Median age using a Geneva-trained
Neural Network (AgeGenevamedian)
269-273 F5.3 Gyr AgeGenevastd Age standard deviation using a
Geneva-trained Neural Network
(AgeGenevastd)
275-280 F6.3 Gyr AgeGenevamode Mode age using a Geneva-trained
Neural Network (AgeGenevamode)
282 I1 --- inDomainGeneva Boolean flag: is within Geneva grid
domain (1=yes,0=no)
(indomainGeneva)
--------------------------------------------------------------------------------
Acknowledgements:
Tristan Boin, tristan.boin(at)obspm.fr
(End) Patricia Vannier [CDS] 12-Feb-2026