This branch of the repository contains notebooks and weights of trained models used for ML4PS workshop submission 69 at NeurIPS 2023. The final paper draft and the poster can be found in the papers section of the workshop website. This work demonstrates likelihood free inference (LFI) on two signal models - a damped harmonic oscillator and a sine-gaussian pulse - whose times of arrival are marginalized using self-supervised learning.
Each directory contains the notebooks and trained weights to reproduce the figures in the paper. In particular, representations in Fig. 2 of the draft are found here for SHO model and here for SG models. Comparison posteriors between a trained flow model with similarity training vs. that without similarity pre-training and nested sampling in Fig. 3 of the draft are found here for the SHO model and here for the SG model.