- Clean and separate the base code in jl file : documentation, using should work, maybe create custom types, etc.
- Systemetize the simulation - obtain many plots with different choices of SNR, generating initial graphs, generating b vector (see the GTF paper for 3 versions)
- Evaluation metrics: PR curve, KL divergence (differential entropy), 2-norm (or infty norm) btw a
- Interesting regimes: sparse graphs, dense graphs, larger graphs (trends in eval metrics over time)
- Datasets: hep citations, wikipedia
- Filling out the model @JS
- Optimization details (derivations in supplement) @AC
- Intro & Related Work @JS
- Write up experiments - later