10:30–11:00 am ERC 401
Megan Zhao "Convolutional Neural Network for Type Ia Supernova Distance Estimation"
With upcoming surveys observing tens of thousands of Type Ia Supernova (SNe Ia), traditional SNe Ia redshift and distance estimations that rely on spectroscopic data will be impractical, making the development of photometric data only estimators necessary. We present a Convolutional Neural Network (CNN) based method for predicting SNe Ia redshift and distance from multi-band supernova lightcurves. We apply our model to simulated Vera C. Rubin Legacy Survey of Space and Time data, which are processed through a 2D Gaussian Process (GP) to create two-dimensional image representations of the lightcurves. Our model achieves satisfactory but limited accuracy, with R-squared scores greater than 0.85. We also present results from a mixed input CNN that allows the addition of scalar values, like peak flux values. This work is the first to use CNN regression in SNe Ia distance estimations and demonstrates that such a model is a viable method for photometric redshift and distance predictions. We discuss the implication of the results and directions for future machine learning based distance estimators.
