J/AJ/157/169 Identifying exoplanets with deep learning in K2 (Dattilo+, 2019)
Identifying exoplanets with deep learning.
II. Two new super-Earths uncovered by a neural network in K2 data.
Dattilo A., Vanderburg A., Shallue C.J., Mayo A.W., Berlind P., Bieryla A.,
Calkins M.L., Esquerdo G.A., Everett M.E., Howell S.B., Latham D.W.,
Scott N.J., Yu L.
<Astron. J., 157, 169 (2019)>
=2019AJ....157..169D 2019AJ....157..169D (SIMBAD/NED BibCode)
ADC_Keywords: Stars, double and multiple ; Stars, dwarfs ; Exoplanets ;
Stars, diameters ; Models
Keywords: planetary systems - planets and satellites: detection
Abstract:
For years, scientists have used data from NASA's Kepler Space Telescope
to look for and discover thousands of transiting exoplanets. In its
extended K2 mission, Kepler observed stars in various regions of the sky
all across the ecliptic plane, and therefore in different galactic
environments. Astronomers want to learn how the populations of exoplanets
are different in these different environments. However, this requires
an automatic and unbiased way to identify exoplanets in these regions
and rule out false-positive signals that mimic transiting planet signals.
We present a method for classifying these exoplanet signals using deep
learning, a class of machine learning algorithms that have become popular
in fields ranging from medical science to linguistics. We modified a neural
network previously used to identify exoplanets in the Kepler field to
be able to identify exoplanets in different K2 campaigns that exist in
a range of galactic environments. We train a convolutional neural network,
called AstroNet-K2, to predict whether a given possible exoplanet signal
is really caused by an exoplanet or a false positive. AstroNet-K2 is highly
successful at classifying exoplanets and false positives, with accuracy of
98% on our test set. It is especially efficient at identifying and culling
false positives, but for now, it still needs human supervision to create
a complete and reliable planet candidate sample. We use AstroNet-K2 to
identify and validate two previously unknown exoplanets. Our method is
a step toward automatically identifying new exoplanets in K2 data and
learning how exoplanet populations depend on their galactic birthplace.
Description:
In this paper we expand on the work of Shallue & Vanderburg
(2018AJ....155...94S 2018AJ....155...94S), which was designed to distinguish planet candidates
and false positives in Kepler data, to classify these signals in data
from the K2 mission. Following Shallue & Vanderburg (2018AJ....155...94S 2018AJ....155...94S),
we use a supervised convolutional neural network architecture, but we make
several key modifications to enhance the network's ability to classify
signals in the qualitatively different K2 data set. Our neural network
utilizes "supervised learning", which means we provide the neural network
with a labeled set of examples from which it can learn. We call this
a training set. Our training set consists of possible planet signals that,
following the naming convention in the literature, we refer to as
"threshold crossing events" or TCEs. These are potentially periodic signals
(decreases in the brightness of a star) that have been detected by an
algorithm designed to search for transiting exoplanets in a light curve.
The total data set used for training comprised 27634 TCEs. We randomly
shuffled and divided the data into three subsets: training (80%,
22105 TCEs), validation (10%, 2774 TCEs), and test (10%, 2755 TCEs).
We used the test set to evaluate final model performance, and we used
the validation set to check performance to optimize the metaparameters.
The random shuffling allows data from each campaign to be spread evenly
among the subsets, and the separate test set allows a cleaner final
performance evaluation because it is not used to train the model or select
the metaparameters. Table 3 includes the training set data for all targets.
File Summary:
--------------------------------------------------------------------------------
FileName Lrecl Records Explanations
--------------------------------------------------------------------------------
ReadMe 80 . This file
table1.dat 79 14 New highly ranked threshold crossing events
(TCEs) from Campaign 12
table3.dat 81 28029 K2 planet candidate training/validation/test set
--------------------------------------------------------------------------------
See also:
IV/34 : K2 Ecliptic Plane Input Catalog (EPIC) (Huber+, 2017)
J/ApJS/222/14 : Planetary candidates from 1st yr K2 mission
(Vanderburg+, 2016)
J/ApJS/226/7 : Planet candidates discovered using K2's 1st yr
(Crossfield+, 2016)
J/MNRAS/465/2634 : Kepler and K2 best candidates for planets (Armstrong+, 2017)
J/AJ/155/21 : Planet candidates from K2 campaigns 5-8 (Petigura+, 2018)
J/AJ/155/119 : HATSouth-K2 C7 transiting/eclipsing systems (Bayliss+, 2018)
J/AJ/155/136 : Planets orbiting bright stars in K2 campaigns 0-10
(Mayo+, 2018)
J/AJ/156/22 : Planetary candidates from K2 Campaign 16 (Yu+, 2018)
J/AJ/156/78 : 44 validated planets from K2 Campaign 10 (Livingston+, 2018)
J/AJ/156/277 : Sixty validated planets from K2 campaigns 5-8
(Livingston+, 2018)
Byte-by-byte Description of file: table1.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 9 I9 --- EPIC [245944045/246414947] EPIC identifier
11 I1 --- TCE [2/6] TCE number
13- 21 F9.6 d Per [0.359275/28.8698] Period
23- 30 F8.3 d T0 [2905.72/2930.72] Epoch of TCE (BJD-2454833)
32- 37 F6.4 h Dur [0.8636/6.2716] Duration of TCE
39- 44 F6.4 --- (Rp/R*)2 [0.0119/0.0828] Square of the planet to stellar
radii ratio (Rp/R*)2
46- 52 F7.5 --- IP [0.0005/1.01823] Impact parameter
54- 58 F5.2 mag Kpmag [9.78/18.71] Kepler magnitude
60- 64 F5.2 --- S/N [9.04/52.32] Signal-to-noise ratio
66- 71 F6.4 --- Pred [0.3197/0.9839] Prediction score
73- 79 A7 --- Note Note(s) (1)
--------------------------------------------------------------------------------
Note (1): Note as follows:
a = Likely instrumental systematic;
b = TCE host star saturated on the Kepler detector, introducing additional
systematics;
c = Validated as K2-294 b;
d = Likely astrophysical signal, but transit shape is reminiscent of an
astrophysical false positive;
e = Strong planet candidate;
f = Validated as K2-293 b;
g = Possible super-Earth/sub-Neptune-sized planet;
h = Possible sub-Saturn-sized planet;
i = G-dwarf host star;
j = M-dwarf host star;
k = K-dwarf host star.
--------------------------------------------------------------------------------
Byte-by-byte Description of file: table3.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 9 I9 --- EPIC [200001049/251809648] EPIC identifier
11 I1 --- Num [1] Planet number
13- 24 F12.9 d Per [0.100011287/63.83604465] Orbital period
26- 36 F11.6 d Time [1977.263977/3338.518579] Time of transit
(BJD-2454833)
38- 50 F13.9 h Dur [0.058113981/131.6147502] Transit duration
52 A1 --- Label Training label (1)
54- 55 I2 --- Camp [1/16] Campaign number
57- 69 F13.9 --- Rp/R* [0.003345256/101.8716318] Planet to stellar radii
ratio
71- 81 E11.5 --- IP [1.18e-06/295.826] Impact parameter
--------------------------------------------------------------------------------
Note (1): Training label as follows:
J = Junk, and typically are either instrumental systematics or stellar
variability;
E = Eclipsing binary star;
C = Planet candidate;
R = Not explained in this paper;
U = Not explained in this paper.
--------------------------------------------------------------------------------
History:
From electronic version of the journal
References:
Shallue et al. Paper I. 2018AJ....155...94S 2018AJ....155...94S
Yu et al. Paper III. 2019AJ....158...25Y 2019AJ....158...25Y
(End) Prepared by [AAS], Tiphaine Pouvreau [CDS] 18-Jul-2019