Importance-weighted ELBO and SUMO (stochastically unbiased marginalization objective) estimator implemented in Jax, applied to a toy problem of approximate sampling from Neal's funnel density.
See associated blog post for more details.
pip install -r requirements.txt
python3 train.py -h