Please enable JavaScript.
Coggle requires JavaScript to display documents.
Andrew Tkanchenko - Stellar astrophysics in the era of big data…
Andrew Tkanchenko - Stellar astrophysics in the era of big data
Stellar evolution
nebula
high mass
main sequence
red supergiant
supernova
neutron star
Black hole
Low mass
main sequence
redgiant
planetary nebula
white dwarf
Global picture - quite well understood
Individual evolutionary phases - large discrepencies between theory and observations
Big progress in 21st century but theory is inferior
95% of lifetime in main sequence and red giant phase
Life determined by interior physics
Starquakes
Probe stellar interiors
maths
Astroseismology
pressure (p-) modes = sound waves
Mostly probe outer layers of stars;radial displacement dominates
Gravity (g-) modes/waves
Probe deep stellar interior;horizontal displacement dominates
Zoo of pulsating stars
Pulsating stars are found everywhere in the HR
Any star will become unstable at least once during its lifetime
Stochastic excitation:convective envelope
Heat-driven pulsations:convective core
evolution of observations: 1-2 modes, 10s modes and short time base, 100s of modes and structure
Gravity waves from space
detection and interpretation
Besides the fundamental properties of stars (M,R, age etc.) and their interior mixin properties, we are able to deduce near-core rotation rates of these stars
g-modes (near core) + p-mode/rotational modulation = core-to evelope rotation
Strong coupling in evolved stars (left) - order of magnitude discrepency with models
Same applies to unevolved stars (right) effect occurs on the main-sequence
TESS
20 million stars per hemesphere
All-sky survey
Many intermediate and high mass stars
1 year of continuous coverage in the CVZs
loads more high mass stars than kepler
AI as a solution
AI - Any technology that enables machines to solve a task in a way like humans do
Machine learning - Algorithms that allow computers to learn from examples with being explicitly programmed
Artificial neural networks - Brain-inspired machine learning models
Deep learning - A subset of ML which uses deep artificial neural networks as models and automatically builds a hierarchy of data representations
AI deep learning
Understanding the world in terms of hierarchy of concepts
Building complicated concepts out of simpler ones
Deep concepts picture with many layers
Linear Regression
Training set -> Learning algorithm -> hypothesis
Algorithms
Deep neural nets - lots of hidden layers
Decision trees - random forest, gradient boosting
Space missions
Kepler to TESS to PLATO
T'DA Pipeline
Photometry
Clustering and watershed using info of the known targets in the image
PSF & difference imaging photometry also implemented but are not currently used
Light curve corrections
ensemble
Construct average weighted light curve from neighbouring targets
Well-suited for long-period/high ampitude variability
CBV approach
Construct sets of Co-trending basis vectors for CCD "areas"
CBVs are split into "slow" and "spike" components
Well suited for high frequency/stochastic/low amplitude variability
Machine learning techniques
sorting out stars into broad classes (light curves only)
Detailed classification (using extra info)
Variability Classification:supervised
Power density spectrum (PDS) -> 128x128 pixels representation of SPD -> Image recognition with neutral nets
Self organising maps on phase curves
Random forest with SOM as a feature
Meta Classifier
Use xgboost as a MetaClassifier
On average gain ~1% overall accuracy compared to best performing individual algorthm
Average accuracy ~97%
TESS: Angular Momentum Transport
All sky, two metallicity regimes
Massive OB stars
young pre-MS stars
Order if magnitude larger samople
Pulsators in eclipsing binaries