Patrick Stokes
Patrick Stokes was a research scientist in the GATAS Lab, where he worked on ASKEM – Automating Scientific Knowledge Extraction and Modeling, the DARPA program on computational modeling with category theory that funded much of the lab’s work on compositional modeling.
He came to the lab from signal processing, statistical modeling and system identification in computational neuroscience. The question there is what a model fitted to a recording does and does not license you to say about the system that produced it, and his paper with Patrick Purdon is the best-known result of that period. It analyses Granger causality, the standard statistical test for information flow between two neural time series, and shows that the way it is ordinarily estimated – fitting a full and a reduced model separately – is either severely biased or, in the variants that remove the bias, badly variable, and that the resulting number does not track the strength of the interaction it is taken to measure. The paper prompted replies from several groups and a sustained exchange in PNAS about what the measure means.
That question is what brought him to category theory: his research interest in the lab was applying it to develop structures and algorithms for robust inference and identification of complex, high-dimensional, multi-scale and stochastic systems.
After the GATAS Lab he went to the GatorSense Machine Learning and Sensing Lab, working with Alina Zare in the UF Department of Electrical and Computer Engineering.
Selected publications
- A study of problems encountered in Granger causality analysis from a neuroscience perspective – Proceedings of the National Academy of Sciences, 2017. Patrick A. Stokes and Patrick L. Purdon. DOI: 10.1073/pnas.1704663114