Rafal Urbaniak

Causal modeling, probabilistic machine learning, and the occasional bit of formal philosophy

⬇ Print / Save as PDF

Rafal Urbaniak

Causal modeling, probabilistic machine learning, and the occasional bit of formal philosophy.

rfl.urbaniak@gmail.com · Google Scholar · GitHub · ORCID · PhilPeople

Research scientist and technical project lead at the Basis Research Institute. I build Bayesian and causal models for high-dimensional, partly observed systems and take them from formal method to production code. Recent projects include circuit discovery in language models, epidemiological model evaluation, spatio-temporal valuation, collective animal behavior, online harm reduction, and the dynamics of policy interventions. I am a core contributor to ChiRho, an open-source causal extension of Pyro/PyTorch, and lead its probabilistic causal impact work. PhD in Logic and Philosophy of Mathematics (Calgary); published in NeurIPS, Computational Linguistics, and Philosophy of Science.

Technical Skills

  • Probabilistic ML & deep learning: Bayesian deep learning, Gaussian processes, normalizing flows, neural ODEs, hierarchical models (PyTorch, JAX, Pyro, NumPyro, ChiRho, GPyTorch, ArviZ)
  • Causal inference: probabilistic explanation & attribution, counterfactual reasoning, actual causality (Halpern–Pearl)
  • Bayesian inference & uncertainty: MCMC, variational inference, uncertainty quantification
  • Dynamical & spatio-temporal modeling: state-space & dynamical-systems models, spatial & spatio-temporal modeling, time series (Dynestyx)
  • Gradient boosting & classical ML: XGBoost, LightGBM, CatBoost, scikit-learn
  • Languages & data: Python (primary), SQL, R; NumPy/SciPy, pandas, geopandas, statsmodels
  • MLOps & infrastructure: Databricks, MLflow, Docker, papermill, Snowflake/DuckDB, RunPod

Positions

  • Research Scientist (Causal Probabilistic Programming), Basis Research Institute, Cambridge, MA2022–
    • Technical project lead on research and client projects.
    • Core contributor to ChiRho. I lead its explainable-reasoning module and the probabilistic causal impact work built on it.
    • Technical lead on circuit discovery in language models via actual causality (arXiv, 2026), on hierarchical Gaussian-process models for spatio-temporal valuation, and on Bayesian models of collective animal behavior (Scientific Reports, 2024).
    • Under the DARPA ASKEM program, co-authored the NeurIPS 2025 counterfactual semantics for hybrid dynamical systems, built its necessity-and-sufficiency tooling in ChiRho, and contributed to tools that extract epidemiological models from the literature.
    • Built Polis, a policy-transfer tool that pairs jurisdiction similarity with causal inference, at The Opportunity Project 2023 sprint (US Department of Commerce). It serves as the backend of a Minneapolis model of how the city's 2015 parking reform affected affordable housing.
  • Technical Lead (contractor), Samurai Labs2019–2022
    • Designed field experiments delivering AI-assisted counter-speech to aggressive Reddit users, and built the Bayesian analysis pipeline behind four journal papers on online aggression.
  • Associate Professor, University of Gdańsk (Philosophy, Sociology & Journalism)2014–2024
    • Principal investigator on two National Science Centre grants applying Bayesian networks and probabilistic modeling to legal and forensic evidence, leading teams of four to eight researchers. The applied-statistics results appeared in Artificial Intelligence and Law and Computational Linguistics.
  • Postdoctoral Fellow (Research Foundation – Flanders, FWO), Centre for Logic and Philosophy of Science, Ghent University (concurrent with Gdańsk)2012–2019
    • Research on logic and the foundations of mathematics, including a Springer monograph and 23 journal articles in logic and philosophy (Synthese, Journal of Philosophical Logic, Studia Logica, Philosophia Mathematica, Erkenntnis, History and Philosophy of Logic, The Review of Symbolic Logic).
  • Assistant Professor, University of Gdańsk2008–2014
  • Postdoctoral Researcher, Centre for Logic and Philosophy of Science, Ghent University (concurrent with Gdańsk)2008–2012

Open-source software

  • ChiRho: causal probabilistic programming on Pyro/PyTorch. Core contributor; lead of the explainable-reasoning module; designed its batched multi-world counterfactual handler.
  • collab-creatures: Bayesian and causal models of group foraging (Scientific Reports, 2024). Lead developer.

Honors & visiting fellowships

  • Kosciuszko Foundation Fellowship, host: B. Fitelson, Northeastern University2022
  • Bednarowski Trust Fellowship, host: M. Smith, University of Edinburgh2022
  • Trinity College Long Room Hub Visiting Fellowship, Dublin2012
  • Visiting Fellow, Banaras Hindu University, Varanasi2011
  • British Academy Visiting Fellowship, University of Bristol2009

Education

  • Habilitation in Philosophy, University of Warsaw2013
  • PhD in Logic and Philosophy of Mathematics, University of Calgary2008

Grants

  • Probability on Trial, OPUS grant, National Science Centre, Poland (PI)2021–2024
  • Forensic & judicial applications of probability, Sonata Bis grant, National Science Centre, Poland (PI)2017–2020

Publications

A Springer monograph, an edited Springer volume, a book under contract with Oxford University Press, a guest-edited special issue, and 40+ journal articles (530+ citations, h-index 12). The full list, grouped by area, is at rfl-urbaniak.github.io/menu/publications.html and on Google Scholar.