Aubrey Barnard

---

I am a computer scientist doing machine learning research, mainly related to medical applications of causal discovery in databases of electronic health records. My research interests include algorithms, causality, probabilistic graphical models, graphs, event history analysis / time series, multi-relational rule learning, and databases.

In 2019, I earned my PhD in Computer Sciences from the University of Wisconsin, advised by David Page (who has moved to Duke). My dissertation was on discovering the adverse effects of medications, through learning the structure of Bayesian network causal models, and through analyzing observational studies with machine learning for hypothesizing drug effects. This research produced a new method for Bayesian network structure learning, and a novel causal discovery machine learning approach based on analyzing before–after studies with temporal inverse probability weighting.

While I mostly work with Python (e.g., scikit-learn, NumPy / SciPy, matplotlib, PyTorch), I have been writing all my numerical code in Julia. It is as easy to use as Python—it is interactive, high-level, expressive, multi-paradigm, dynamically-typed—but it runs at machine speed and has linear algebra and concurrency built in. I encourage you to check Julia out!

I ran the UW–Madison ML and AI Reading Group for 5 semesters.

---

Research Interests

• Algorithms
• Causality (in observational data)
• Probabilistic graphical models (including structural causal models), structure learning, inference
• Graph theory & algorithms
• Event sequences / time series (patient histories in electronic health records can be modeled as irregular, sparse, and noisy sequences of events)
• Multi-relational rule learning (inductive logic programming)
• Databases

Other Interests

• Open source software
• Software design and development
• Programming languages
• Linux
• Go (the game, but the language is cool, too)

Current Research Projects

My research approaches machine learning from a computer science perspective, focusing on improving efficiency through new algorithms or mathematical insights, or sometimes just filling in gaps. I have research in progress on the following:

• Identifying ovarian cancer earlier by inspecting electronic health records
• Pairwise interactions are sufficient for independence testing; generalized Hammersley–Clifford theorem
• Non-combinatorial Bayesian network structure learning via convex optimization
• Efficiently enumerating relevant cycles
• Scalable matching
• Speeding up cross validation with experimental design
• Principled, statistical comparison of graphs for evaluating structure learning
• Any-time inference for log-linear Markov networks via decreasing likelihood enumeration
• Better optimization for fitting log-linear models
• Faster and more optimal inductive logic programming via frequent itemset mining
• Replacing noisy-OR

Selected Papers

• Pairwise Interactions are Sufficient for Independence Testing
  Aubrey Barnard, Scott Alfeld
  In preparation

• Temporal Inverse Probability Weighting for Causal Discovery in Controlled Before–After Studies: Discovering ADEs in Generics
  Aubrey Barnard, Peggy Peissig, David Page
  Causal Learning and Reasoning 4 (PMLR 275), 2025
  paper, poster, bibtex

• Causal Discovery of Adverse Drug Events in Observational Data
  Aubrey Barnard
  PhD Dissertation, Computer Sciences, University of Wisconsin–Madison, 2019
  bibtex

• Causal Structure Learning via Temporal Markov Networks
  Aubrey Barnard, David Page
  Probabilistic Graphical Models 9 (PMLR 72), 2018
  combined paper & supplement (NLM version), poster, bibtex

• Identifying Adverse Drug Events by Relational Learning
  David Page, Vítor Santos Costa, Sriraam Natarajan, Aubrey Barnard, Peggy Peissig, Michael Caldwell
  AAAI 26, 2012
  bibtex

Google Scholar Profile