Research & Applied Work
Kurt Pasque
Decision Boundaries · Robust ML · Neural Reasoning
I study how machine learning models draw decision boundaries, why those boundaries fail, and how to make them more robust. My work uses tropical geometry to understand and build neural networks, with experiments that test what holds up under noise, adversarial inputs, and unfamiliar data.
I want to build models that can reason across different kinds of sensor data and support decisions in the physical world. Defense and critical infrastructure are where I want to put that research to work. I bring a background in operations research, hands-on model development, and experience applying AI inside the Navy.
U.S. Navy officer · Data Scientist, Naval Education and Training Command · Research Collaborator, Naval Postgraduate School
Current Focus
-
Decision Boundaries & Robustness
Understanding the geometry behind a model's decisions and testing when small changes to an input cause it to fail.
-
Neural Algorithmic Reasoning
Building attention mechanisms that reflect the structure of a problem, then testing whether they generalize beyond their training conditions.
-
Reasoning Across Sensors
The direction I want to take this work: custom architectures that combine different sensor modalities and remain useful when observations are noisy or incomplete.
-
Decision Systems
Using AI and operations research to help people prioritize resources under real budget and mission constraints.
Selected Research
-
Phylogenetic Tree Inference with Tropical Axial Attention
Accepted August 6, 2026
Tropical axial attention for learning phylogenetic distance structure from multi-species sequence alignments. Code · Preprint.
-
Tropical Attention: Neural Algorithmic Reasoning for Combinatorial Algorithms
Introduces tropical attention as a structured alternative to standard attention for neural algorithmic reasoning. Code.
-
Adversarially robust neural network decision boundaries via tropical geometry
Connects tropical-geometric structure in neural decision boundaries with empirical adversarial robustness.
-
Tropical Geometry in Neural Networks: A Novel Defense against Adversarial Attack
Foundational thesis work on tropical representations of neural networks and adversarial defense.
Selected Work
-
Challenge Winner · Individual Effort · 2026
Human Capital Optimization Challenge
I built an LLM analysis workflow using GenAI.mil inside the War Data Platform to assess the marginal strategic value of more than 6,000 tags applied to Navy billets. The aim was practical: give decision-makers a better basis for prioritizing manpower requirements within a constrained budget.
This individual effort won the directorate's Human Capital Optimization Challenge. I received the award at AFCEA East in spring 2026 and briefed the approach to several organizations exploring how to integrate AI into their work.
Award presentation · AFCEA East, spring 2026 -
Tropical Attention
Co-developed the research and contributed substantial implementation work to the official Tropical Attention codebase. The project studies how tropical geometry can shape neural reasoning architectures and their robustness. Repository · Baran-phys/Tropical-Attention.
-
Universal Reasoning Models
Experimental work on recursive reasoning architectures and tropical-attention variants for challenging reasoning benchmarks. Repository.
-
Tropical Axial Attention Reproduction Pipeline
End-to-end research code for simulation, model inference, tree construction, evaluation, and experiment aggregation in neural phylogenetics. Repository.
-
Eval First Draft + FitRep First Draft
Built and released lightweight AI-assisted drafting tools with front-end workflows, API-backed generation pipelines, and iterative reliability improvements from user feedback. Eval First Draft · FitRep First Draft.
-
Naval Postgraduate School Return on Investment Analysis
Developed quantitative metrics to communicate educational ROI to Department of Defense stakeholders. Repository.
Background
Experience
-
Data Scientist · Naval Education and Training Command
Leads AI integration programs, training modernization efforts, and data products supporting senior-level decisions.
-
Research Collaborator · Naval Postgraduate School
Research in tropical geometry, neural algorithmic reasoning, and experimental reasoning architectures.
-
Division Officer · U.S. Navy Pay and Personnel Support Center
Built UiPath automations for back-office workflows and led transition activities from legacy systems to Salesforce CRM.
Education
-
Naval Postgraduate School · M.S., Operations Research
-
United States Naval Academy · B.S., Mechanical Engineering
Recognition & Certifications
-
Human Capital Optimization Challenge Winner
ASN M&RA MAHRS directorate · Individual effort · Award presented at AFCEA East, spring 2026. Project & award photo.
-
Research Recognition
INFORMS Military & Security Society Donald P. Gaver, Jr. Thesis Research Award winner (2025) · MORS Stephen A. Tisdale Graduate Research Prize finalist.
-
Professional Credentials
Certified Analytics Professional-Expert · Professional in Human Resources · UiPath Certified Advanced RPA Developer · Fundamentals of Engineering Exam.