J/MNRAS/476/3233 14983 Kepler red giants (Hon+, 2018)
Deep learning classification in asteroseismology using an improved neural
network: results on 15 000 Kepler red giants and applications to K2 and
TESS data.
Hon M., Stello D., Yu J.
<Mon. Not. R. Astron. Soc., 476, 3233-3244 (2018)>
=2018MNRAS.476.3233H 2018MNRAS.476.3233H (SIMBAD/NED BibCode)
ADC_Keywords: Stars, giant ; Optical ; Asteroseismology
Keywords: asteroseismology - methods: data analysis - stars: evolution -
stars: oscillations - stars: statistics
Abstract:
Deep learning in the form of 1D convolutional neural networks have
previously been shown to be capable of efficiently classifying the
evolutionary state of oscillating red giants into red giant branch
stars and helium-core burning stars by recognizing visual features in
their asteroseismic frequency spectra. We elaborate further on the
deep learning method by developing an improved convolutional neural
network classifier. To make our method useful for current and future
space missions such as K2, TESS, and PLATO, we train classifiers that
are able to classify the evolutionary states of lower frequency
resolution spectra expected from these missions. Additionally, we
provide new classifications for 8633 Kepler red giants, out of which
426 have previously not been classified using asteroseismology. This
brings the total to 14983 Kepler red giants classified with our new
neural network. We also verify that our classifiers are remarkably
robust to suboptimal data, including low signal-to-noise and incorrect
training truth labels.
Description:
We have developed a variant of our previous deep learning classifier,
which improves the prediction accuracy of our previous classifier and
avoids predicting HeB stars at high {DELTA}ν. We trained improved
classifiers for 4-yr, 356-d, 82-d, and 27-d photometric time series,
which are representative for Kepler, K2, TESS, and large parts of the
PLATO sample. In order to optimize the classifier's performance, we
determined the probability threshold pthres that maximizes the
classifier accuracy. As a result, we could report test set accuracies
of 98.3 per cent (4yr), 98.3 per cent (356d), 95.4 per cent (82d), and
93.2 per cent (27d).
File Summary:
--------------------------------------------------------------------------------
FileName Lrecl Records Explanations
--------------------------------------------------------------------------------
ReadMe 80 . This file
table3.dat 53 14983 Population class predictions for
14983 Kepler red giants
--------------------------------------------------------------------------------
See also:
V/133 : Kepler Input Catalog (Kepler Mission Team, 2009)
J/MNRAS/469/4578 : Deep learning classification in asteroseismology
(Hon+, 2017)
Byte-by-byte Description of file: table3.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 8 I8 --- KIC KIC number
10 I1 --- Pop [0/1] Population (1)
12- 16 F5.3 --- Prob ?=-99 Probability (2)
18- 22 F5.3 --- e_Prob ?=-99 Uncertainty on probability
24- 31 F8.5 --- deltanu Mean large frequency separation of modes with
the same degree and consecutive order,
{DELTA}nu (Yu et al., in prep.)
33- 41 F9.5 --- numax Frequency of maximum oscillation power
(Yu et al., in prep.)
43- 49 F7.3 --- epsilon ?=-99 Location of the l=0 mode (3)
51- 53 A3 --- Flags Flags (4)
--------------------------------------------------------------------------------
Note (1): Population as follows:
0 = RGB
1 = HeB
Note (2): The threshold is p=0.567.
Stars with probabilities above this are predicted HeB.
Note (3): epsilon=1/4+alpha, where alpha is the contribution from the outer
turning point, which is determined by the properties of the near-surface
region of the star (see Huber et al. (2010ApJ...723.1607H 2010ApJ...723.1607H), Sect. 3.3).
Values of epsilon = -99 indicate that epsilon was not provided.
Note (4): Flags as follows:
U = Star is in the 'unclassified set', which is the set of 8633 stars
without asymptotic period spacing measurements
R = Star is used in the training set in Hon et al. (2017MNRAS.469.4578H 2017MNRAS.469.4578H,
Cat. J/MNRAS/469/4578). Predictions are from 10-fold cross validation,
in particular, when the star is in the validation fold
T = Star is used in the test set in Hon et al. (2017MNRAS.469.4578H 2017MNRAS.469.4578H,
Cat. J/MNRAS/469/4578)
V = Deep learning prediction disputed by Mosser/Vrard
E = Deep learning prediction disputed by Elsworth et al.
(2017MNRAS.466.3344E 2017MNRAS.466.3344E, Cat. J/MNRAS/466/3344)
S = Deep learning prediction disputed by Stello et al.
(2013ApJ...765L..41S 2013ApJ...765L..41S, Cat. J/ApJ/765/L41)
N = HeB prediction at numax>115uHz, a range where HeB stars should not exist.
A likely incorrect prediction.
--------------------------------------------------------------------------------
History:
From electronic version of the journal
(End) Patricia Vannier [CDS] 21-Sep-2023