I am a statistician and applied mathematician with broad interests in
statistical methodology and computing in the physical sciences. My primary
interest is modeling complicated dependence structure in real-life
temporal/spatial/spatio-temporal processes, and I'm particularly interested in
theoretical questions that are motivated by computationally scalable methods and
approximations.
Education:
Ph.D. Statistics, Rutgers University (2018 – 2023)
Dissertation: Models and Computation for Nonstationary Processes at Scale
Adv. Michael L. Stein
B.S. Mathematics / B.A. Statistics, University of Chicago (2012 – 2016)
Employment:
University of Wisconsin-Madison, Dept. of Statistics
Assistant Professor (2023 – )
Argonne National Laboratory, Division of Mathematics and Computer Science
Clicking on the title of the paper will take you to ArXiv or some other direct
PDF source, and the DOI link will take you to the official journal version.
Direct PDF links marked (ArXiv vXXX) mean that the ArXiv version differs from
the journal version somehow. I try to do a good job about uploading substantial
revisions to arXiv, so the differences are in general limited to small revisions
that don't really change the substance of the paper very much. But if you want a
journal version and can't get access, please shoot me a note and I'll send one
over.
All code is written in the Julia language. If you are an R user but want to use
any of this software, feel free to reach out---I have provided or helped other
people develop bindings for use in R before. It isn't too bad to do.
SMultitaper.jl:
a bare-bones adaptively-weighted multitaper spectral density estimator.
Makes aggressive use of caching for fast and non-allocating repeated
calls.
GPMaxlik.jl:
A backend library for plugging in your exotic implicit matrix operator and
getting a careful and efficient Gaussian negative log-likelihood, score,
and expected Fisher information matrix. I actually wrote docs for this
one, so bring your FMM code and skip coding the GP parts by using this
library.
Vecchia.jl,
A chunked implementation of Vecchia's approximation. Carefully tuned for
great thread-parallelism and automatic differentiation. This software is a
testbed for a lot of what I do and offers a lot of functionality that no
other scalable GP library provides, so please do give it a shot.
BesselK.jl,
An implementation of the modified second-kind Bessel function K_v that is
`ForwardDiff.jl`-compatible with respect to the order. See the paper
"Fitting Matern..." for more details on this.
libmatern,
An improved version of the besselk routines in BesselK.jl, and now written
in C for portability. Hook into it with your favorite programming
language and existing software stack!
SpectralKernels.jl,
a library for evaluating covariance functions in one dimension via their
spectral density. It is the software companion to "Fast Adaptive
Fourier...", and has a lot of neat tricks for scaling up well and handling
difficult spectral densities, like with singularities or slow tail decay.
It is also designed with autodiff in mind, so you can automatically get
gradients of your spectrally-specified models as well.
SpectralEstimators.jl,
The software companion to "Fast Machine-Precision...", gives exact
log-likelihoods for time series models with piecewise smooth spectral
densities in quasilinear time. I haven't gotten around to writing much
documentation for this, but if you are interested in hooking in, shoot me
a note and I'll develop whatever docs you need.
Contact:
Besides for a very new Mastodon account
that I will probably not actually get into the habit of using, I do not use any
social media or internet networking websites. If you see any account with my
name on it, please assume that it is not really me. Please contact me by email
at the following address:
geoga $at wisc $dot edu
I also use XMPP, so if you'd prefer to communicate by instant messaging please
first reach out by email and I will share that contact information with you.