Weak Lensing Cosmology with Machine Learning

For my Master's, I explored Deep Learning as a method of constraining cosmological parameters in the Lambda-Cold-Dark-Matter Model, supervised by Benjamin Giblin and Catherine Heymans. We used noisy weak lensing shear maps generated using the SLICS and CosmoSLICS simulation suites to train a convolutional neural network (CNN) to predict the underlying parameters. During this process, we explored the effects of different network architectures, training methods, and data augmentation techniques. We found that a CNN was not yet competetive with the standard 2-point correlation function method, but that it was able to obtain constraints on the cosmological parameters explored.