Based on the research done by Anne Dattilo, a Ph.D. student in Astronomy and Astrophysics at the University of California Santa Cruz.
What are Exoplanets??
Planets that are outside our solar system are called exoplanets, these planets orbit any star other than the sun. You will be astonished to know that there are 4000 exoplanets have already been discovered via methods like radial velocity, transits, direct imaging, microlensing out of which transits is the most prolific and used for this project.
What is Neural Network and How it is used here?
In this research Convolution, Neural network (CNN) is used which is an artificial neural network, a class of deep neural networks. CNN is commonly and effectively used to analyze visual imagery.
CNN generally consists of – Convolution Layer, Non- Linear activation function, pooling layer, Fully- connected network.
For a more detailed explanation on CNN like what happens in these layers? How does CNN learn the features and classifies the images? Stay tuned for an elaborate blog on neural networks.
If you are new to neural networks, do checkout Understanding Neural Networks
In this research, CNN is used to find planets in the K2 data which is data collected by Nasa’s dedicated Kepler space telescopes.
Data preprocessing, splitting, and Training the model
The brightness plotted overtime of a star is called as a light curve. If a planet were to pass in front of its host star, the light we receive from the star dim and there we get a dip in our light curve which is called a transit event. The presence of the planet is found if this happens periodically. 51,711 signals, 31,575 of which were classified into three categories was used for training the neural network. CNN works best when the data are of similar shape and size, but phase-folded light curves can come in many shapes and sizes based on the characteristics of the planetary system. The orbital period and transit depth of planets differ. So, the data was processed into two image features to have the planets with the same shape and size, which were local view and global view where the global view is the entire phase-folded light curve with the transit event in the center and a local view is a zoomed-in look at the transit event, with only two durations on either side of the event instead of the entire period. The features were normalized to keep the transit depth as -1.
The CNN was trained on 80% of the signals using the Adam optimizer for 4000 Epochs, 10% for validation, and 10% for our final test set.
After the Training, the model was tested with the testing set to see if it had learned what a planet is based on the predictions for each signal to use it on a new set of data to find new planets. The model achieved a 98% accuracy over the test set and was used to identify and verify two new exoplanets.
Stay Tuned for more freaky news and facts. Feel free to share your thoughts in the comment section and do check our other posts on cyber and AI here.
Follow us on Instagram for the latest updates in Cybersecurity and AI.
You can find me on LinkedIn
Comments