Madelyn Andersen, MSc

Madelyn Andersen, MSc

she/her

I’m a PhD student in Bayesian Statistics in Tamara Broderick's group within LIDS at MIT, working on variational inference, robustness, and probabilistic modeling.

Guide

What this page is

This is a lightweight guide to what I’m working on, how my projects fit together, and the best way to engage with my work.

Research snapshot

My research focuses on improving variational inference (VI) methods—especially the practical reliability of approximate Bayesian inference in modern probabilistic models. I’m interested in how default choices (parameterizations, transformations, optimizers, and initialization) shape convergence behavior and posterior approximation quality.

Current themes

  • Robust variational inference: making VI more reliable across model classes and parameterizations.
  • Default settings & diagnostics: understanding when “reasonable defaults” fail and how to detect it early.
  • Posterior approximation quality: clearer metrics for comparing approximations beyond a single scalar objective.
  • Probabilistic programming workflows: reproducible benchmarking across frameworks and implementations.

Start here

If you’re new to my work, the best entry points are:

How I work

I care a lot about:

  • reproducibility (seed control, clean configs, tracked environment details),
  • comparability (consistent evaluation across methods),
  • clarity (simple baselines and transparent assumptions).

Collaboration

I’m excited to collaborate on projects related to variational inference, probabilistic modeling, and robust ML. If you’d like to reach out, email is best.