Data reduction, emulation and inference using a combination of GP processes and Variational autoencoder. Currently applied for CMB angular power spectra C_l and matter power spectra P(k). In principle, this can be extended to image emulations as well.
Parameter inference is done using MCMC for cosmological parameters, with public PLANCK/WMAP/SPT data.
scp phoenix:/homes/nramachandra/AstroVAE/Cl_data/Model/7500 mcs:/homes/nramachandra/DataP5/Model/ scp phoenix:/homes/nramachandra/AstroVAE/Cl_data/Data/norm7500 mcs:/homes/nramachandra/DataP5/ scp phoenix:/homes/nramachandra/AstroVAE/Cl_data/Data/mean7500 mcs:/homes/nramachandra/DataP5/ scp phoenix:/homes/nramachandra/AstroVAE/Cl_data/Data/encoded7500 mcs:/homes/nramachandra/DataP5/
scp P25. mcs:/homes/nramachandra/DataP5/raw/
- Error propoagation using Bayesian neural networks
- t-SNE for reduction and visualization
- RNN for time analysis
- Adversarial AEs