We develop machine-learning and data-science methods for finding, interpreting, and using structure in complex, high-dimensional data. Our work connects mathematics, statistics, machine learning, signal processing, and information theory, with applications in biology and medicine, remote sensing, environmental science, navigation, and national security.
The lab is directed by Kevin R. Moon, Associate Professor in the Department of Mathematics and Statistics and Founding Director of the Data Science and Artificial Intelligence Center at Utah State University.
Academic website · Research overview · Publications · Google Scholar · USU DSAI Center
- Geometry and representation learning. We develop manifold-learning, supervised-representation, and information-geometric methods that reveal task-relevant structure in complex data.
- Interpretable and trustworthy AI. We study variable importance, higher-order interactions, symmetry discovery and enforcement, robust learning, and methods for understanding what a model has learned.
- Scientific and engineering applications. We apply and extend machine learning in biology and medicine, remote sensing, environmental science, navigation, and national-security problems.
| Project | What it provides | Related research | Status |
|---|---|---|---|
| SymmetryML | Continuous symmetry discovery and enforcement using learned vector fields | NeurIPS 2024 and continuing work | Active research code |
| GRAE | Geometry-regularized autoencoders for extendable and invertible manifold learning | IEEE TPAMI, 2023 | Stable research code |
| RF-GAP | Geometry- and accuracy-preserving random-forest proximities in R | IEEE TPAMI, 2023 | Stable research code |
| FIG | Functional Information Geometry for noisy, high-dimensional dynamical data | SampTA 2025 | Research code |
| PF-GAP | Proximity-based learning for time series, including prediction, imputation, outlier detection, and visualization | ACDSA, 2026 | Active research code |
| Nexarag | Local, graph-based context construction and visualization for transparent and reproducible LLM research | Journal of Open Source Software, 2026 | Active |
- PHATE — state of the art dimensionality reduction method for visualization, maintained by the Krishnaswamy Lab
- RF-PHATE — supervised dimensionality reduction for visualization using random-forest geometry
- CLIQUE — a new variable importance method that mitigates known issues of SHAP and LIME
- EnDive — ensemble estimation of information divergences
- DIG — Dynamical Information Geometry for visualizing data from dynamical processes
Each project README provides installation, examples, and the corresponding paper or citation. When using a method in academic work, please cite the linked paper and the software release when one is provided. Questions and reproducible bug reports are best submitted through the issue tracker of the relevant repository.
Research software in this organization often grows from student-led scholarship. We aim to make published methods inspectable, reproducible, and reusable while giving students experience developing software for researchers outside the group.
See the lab People page for current students and alumni. Prospective students and researchers interested in collaboration can find research descriptions and contact information on the academic website.
Profile updated September 2026.
