jtamanas/saxbi

JAX implementation of Sequential Neural Likelihood Estimation (SNLE) and Sequential Neural Ratio Estimation (SNRE) simulation-based inference algorithms

Hey there superstar 🌟

As a member of both Stefano Profumo’s SCIPP Theory group at UC Santa Cruz and Joseph Hennawi’s ENIGMA group at UC Santa Barbara and Leiden Observatory, I work at the intersection of (high energy/astro)physics, deep learning, and statistics.

In my time as a graduate student, I’ve been lucky enough to work on projects with the world’s leading experts on applied deep learning. I have worked on deepening (pun intended) the understanding of current theories for a variety of topics: From beyond standard model particle physics, to the structure of the Milky Way, and the cosmological history of the universe.

You might know me from my work on:

1. Simulation-Based Inference: Approximate Sequential Bayesian algorithms written in JAX. Used with applications to high energy physics phenomenology and future quasar inference work.
2. Via Machinae: Unsupervised anomaly detection to discover stellar streams in the Milky Way.
3. Spectre: Approximate Bayesian algorithm for SOTA quasar continuum inference.

saxbi's Issues

mask tile

FYI,

In made.py following the comment on tile jax numpy the mask should be update accordingly.

        self.layers = [
            MaskedDense(features=masks[0].shape[-1], mask=masks[0], use_context=True),
            getattr(nn, self.act),
            MaskedDense(features=masks[1].shape[-1], mask=masks[1]),
            getattr(nn, self.act),
            MaskedDense(
                features=masks[2].shape[-1] * self.output_dim_multiplier,
                mask=np.tile(masks[2],self.output_dim_multiplier),    # update <<<<<<<<<
            ),
        ]