Abstract
The work is about identifying planet candidates using deep learning models. Finding these objects manually is a very labor-intensive task. For example, The Large Synoptic Survey Telescope (LSST) is expected to generate about 200, 000 images per year, which is equivalent of more than 106 GB of data. Therefore, using reliable algorithms to manage the data is necessary. Deep learning can be helpful because it suits well for very large input data. In general, having more data only makes deep learning models perform better.
We use Kepler Space Telescope and Transiting Exoplanet Survey Satellite (TESS) to detect planet candidates by using a convolutional neural network model. We apply the Q1-Q17 (DR24) table as our training and test sets. The model takes two phase folded light curves and some parameters of each transit-like signal and then outputs whether the signal represents a planet candidate (PC) a non-transiting phenomena (NTP) or a false positive (FP). In the current model, we feed 17 features into a dense neural network model, such as transit durations and depth of signals. At this stage, the model achieves AUROC and accuracy of about 97.7%, 95.9% respectively for the test set. The accuracy for the training set can be over 99%, which means that the model can easily overfit the data. The most straightforward way to the problem is to use more data to train the model. Therefore, we plan to train it with more simulated data later in order to increase the AUROC and accuracy of predictions.
We use Kepler Space Telescope and Transiting Exoplanet Survey Satellite (TESS) to detect planet candidates by using a convolutional neural network model. We apply the Q1-Q17 (DR24) table as our training and test sets. The model takes two phase folded light curves and some parameters of each transit-like signal and then outputs whether the signal represents a planet candidate (PC) a non-transiting phenomena (NTP) or a false positive (FP). In the current model, we feed 17 features into a dense neural network model, such as transit durations and depth of signals. At this stage, the model achieves AUROC and accuracy of about 97.7%, 95.9% respectively for the test set. The accuracy for the training set can be over 99%, which means that the model can easily overfit the data. The most straightforward way to the problem is to use more data to train the model. Therefore, we plan to train it with more simulated data later in order to increase the AUROC and accuracy of predictions.
Anyone interested is welcome to attend.