02 / Academics

Reason carefully.
Make methods usable.

Statistical methodology, computational optimization, and open research software—built so the assumptions, implementation, and evidence can be inspected.

Google Scholar ↗

02APhD Dissertation

Topics in Trend Filtering with Poisson Loss

Count data appear everywhere—from website traffic and disease surveillance to medical imaging—but their variability grows with magnitude, making constant-noise smoothing unreliable. This dissertation develops Poisson trend filtering methods that adapt to abrupt changes without prespecifying where they occur, together with stable and scalable algorithms for complex data structures and ill-conditioned systems. The methods are demonstrated through applications in epidemic surveillance and spatial analysis.

The question: How can we recover locally adaptive trends from count data without distorting low-count regions or sacrificing computational stability?
The artifact: A PhD dissertation unifying statistical theory, methodology, optimization algorithms, scientific software, and applications to epidemic surveillance and spatial data.

Defence slides ↗Dissertation link forthcoming

Poisson ModelsTrend FilteringConvex OptimizationNumerical Linear AlgebraEpidemic SurveillanceStatistical Theory

02BMethodology

Estimating a time-varying reproduction number

A collaborative R package and research implementation for estimating time-varying effective reproduction numbers using a Poisson likelihood, trend-filtering regularization, and cross-validation.

The question: How can epidemic growth be estimated without forcing the underlying trajectory to be globally smooth?
The artifact: Inspect the public R package, examples, and linked research implementation.

Package site ↗Source code ↗Paper ↗

RC++Convex OptimizationNonparametric regressionPoisson modelComputational biology

02CAlgorithm

Trend filtering, built for use

Open-source implementations of locally adaptive nonparametric regression, including a Python interface with a C++ backend, cross-validation, and structured linear-system solvers.

The question: How do we make flexible statistical estimation computationally practical and reusable?
The artifact: Explore the public R and Python implementations.

Python package ↗R package ↗Source code ↗

PythonRC++Convex optimizationLarge-scale linear system

02DReflection

Ten years of thinking statistically

A long-form reflection on how ten years of studying and practicing statistics changed the way I approach evidence, uncertainty, computation, and the limits of what data can tell us.

The question: What remains after a decade of learning how to reason with data?
The artifact: A personal essay connecting statistical training, research practice, and the habits of mind that extend beyond any single method.

Essay forthcoming

Statistical reasoningResearch practiceScientific writing