Home

Nathan Wycoff, in France in 2018, taking waste disposal seriously.

In Brief: I am an Assistant Professor in the Department of Mathematics and Statistics at the University of Massachusetts Amherst.

My expertise is in Surrogate Modeling, the use of statistical models to understand complex computational models and bio/physical systems, and I am interested more generally in the nexus of ML/AI and optimization, including Bayesian Optimization, Transfer Learning, and Uncertainty Quantification.

Currently, I have three sponsored research projects:

How can we use regression trees to make physics-informed machine learning fast and cheap enough for low-compute edge devices? This is the subject of my 2026 Air Force Young Investigator Program (YIP) award.
How can we do sequential design of experiments in infinite dimensional spaces, such as function spaces? This is motivated by exciting work in designing wearable devices to enhance human mobility by my awesome collaborators at UMass.
How can we do transfer learning with extremely limited and questionably relevant source datasets? Motivated by important work predicting the course of emerging humanitarian crises with great collaborators Georgetown and UNHCR.

I'm very grateful to the Air Force Office of Scientific Research, the National Geospatial-Intelligence Agency, and the National Science Foundation for supporting our work.

email: nwycoff@umass.edu

Selected Research Documents:

(For a full list, see my Google Scholar)

Aditya Kulkarni, Davide Sartor, Carlos Soto, and Nathan Wycoff (2027+). How low can you go: Tiny LoRA LLM Bayesian Optimization (LIMBO) Finetuning. Under Review. <arXiv>
Davide Sartor, Meghan Huber, Donghyun Kim, and Nathan Wycoff (2026). Bayesian Optimization on Function Spaces via Sparse RKHS Manifolds. To Appear in Neurips '26. <arXiv>
Poorbita Kundu and Nathan Wycoff (2026). Active Subspaces in Infinite Dimension. AISTATS '26, <AISTATS> <arXiv> <Github>.
Nathan Wycoff, Ali Arab, and Lisa O. Singh (2026+). Formal Bayesian Transfer Learning via the Total Risk Prior, Under Review, <arXiv>.
Nathan Wycoff (2025). Regression Trees Know Calculus, Neurips '25, <Neurips> <arXiv>.

Software:

- R package activegp: Design for and implementation of black-box sensitivity analysis using Gaussian Processes and the Active Subspace Method.

To get started, just do:

R> install.packages("activegp")

R>example(activegp)

Previous Work Experience:

Postdoc at Georgetown's Massive Data Institute in the McCourt School of Public Policy (Summer 2021-Summer 2024).
Givens Scholar at Argonne National Laboratory / Math and Computer Science Division (Summer 2018).
Web Dev Intern at General Dynamics Mission Systems (Summer 2016).
Business Intelligence Intern at Comprehensive Health Services (Summer 2015).
Intern for the late Congressman Don Young (Spring 2013).

Google Sites

Report abuse