Megan is an excellent undergraduate student who joined my group for the 2025 NSF REU program at Stony Brook and has continued to work with me.

Ms. Schlogl’s REU problem was challenging. In the manuscript arXiv:2505.22574, we investigated how machine learning based on the cutting-edge transformer architecture could accelerate the computation of Cosmic Microwave Background (CMB) temperature and polarization power spectra. Amritpal Nijjar’s thesis served as the basis for Appendix F, where we sought to improve Prof. Wayne Hu’s analytical formula for the unlensed damping tail of the CMB, discovered approximately 25 years ago, to account for the effects of gravitational lensing. These solutions could be used to preprocess the data vectors used by Machine Learning techniques, reducing the size of the training set required to achieve a desired accuracy. 

During her REU summer, she decided to take another simpler approach (compared to Genetic Algorithms), following the Steps of Prof. Hu and attempting to fit polynomial expansions of the cosmological parameters. Making that work is quite laborious, but she prevailed and achieved results that are really good. We are going to publish her results in an upcoming paper about Machine Learning optimizations for beyond LCDM cosmologies. All in just 10 weeks. 

She is now working on machine-learning methods to transfer learn emulators from one model to another (for example, emulators trained on Prof. Hu's analytical solution to models trained on CAMB output).