talks

2023

1. MORSS

   Uncertain Waters: Integrated Modeling of Fuel-Constrained Operations in Contested Environments

   McCabe, L. H., Brown, S., Whittle, S., Danneman, N., Cheng, B., and Anderson, M.

   MORS 91st Symposium 2023

   Abs HTML

   The evolving battlespace has underscored a critical need for de-risked energy supply lines, which warrant innovation in energy demand, fuel production, and contested logistics. Toward the latter end, we introduce COLOGEN (COntested LOGistics ENgine), a library for synthesizing information about the global contestedness landscape and solving constrained routing problems using graph algorithms, and FuelSim, an agent-based simulation application for analyzing fuel-constrained operations and alternative fuel paradigms. Our integration of these software components provides a robust framework for modeling and simulation of contested maritime logistics, where increased attention has been paid to the operational feasibility of sustainable fuels. We illustrate the coherence of our framework via example, focusing on naval exercises in the Pacific.

2022

1. DARPA

   Network Contagion Source Localization

   McCabe, L. H.

   DARPA Forward 2022

   Abs HTML

   Core objective: estimate the origin of information in digital media, but this task is limited by complexity at scale. “Information spread” (broadly defined) can be modeled as diffusion processes on graphs: search for the “patient(s) zero” of an epidemic, whether natural contagion or biological weapon, estimate origin of information propagating across digital media or cyber networks, and compare the likelihood of hypotheses about malware in computer networks generated by IC analysts. Standard setup: undirected “substrate graph‘’ "G=(V, E)" and a source set S⊆V representing the originator(s) of stochastic diffusion and estimate S from some information about the state of the diffusion process at an arbitrary time (e.g. “snapshot” observation).

2. MORSS

   Modeling the Operational Feasibility of Synfuel from Seawater

   Brown, S., Horio, B., Anderson, M., Whittle, S., McCabe, L. H., Johnson, C., and Funk, S.

   MORS 90th Symposium 2022

   Abs HTML

   A critical factor for deterrence is the ability to resupply our forces to maintain extended physical presence and to support sustained military operations. Given continued National interest in the Pacific, future marine and aviation fuel generated from alternative sources such as seawater, is a potential game changer for naval replenishment logistics. In addition to the engineering challenges for making this technology a reality, the impact on logistics and operations must be fully understood. In this talk, we discuss how modeling and simulation may be used to support feasibility studies of sustainable aviation fuel.

2021

1. NeurIPS

   URChIn and Variants (Unsupervised Representations for Chess Inference)

   McCabe, L. H.

   Neural Information Processing Systems 2021

   Abs HTML

   We compare several variations of our core Reconnaissance Blind Chess algorithm, which employs a language model to estimate a plausible state distribution, rather than sampling uniformly over the true state space. We provide experimental evidence that this approach enables multi-hypothesis tracking agents to employ more optimistic action decision rules.

2. SMC

   Geodesic and Neural Features for Link Prediction in the COVID-19 Biomedical Knowledge Graph

   McCabe, L. H.

   Smoky Mountains Computational Sciences & Engineering Conference 2021

   Abs PDF Code Poster

   Motivated by Challenge 2 of the 5th Annual Smoky Mountains Computational Sciences Data Challenge, we analyze the COVID-19 biomedical knowledge graph. After computing geodesic statistics for all nodes in the network, we present several machine learning pipelines for automated hypothesis generation. Our most performant approach achieves classification results comparable to the state-of-the-art on thematically similar link prediction benchmarks. Relevant source code and data is available in our public GitHub repository, described in Section 5.

3. Networks

   Preferential Attachment Estimation and Node Immunization Along Random Walks

   McCabe, L. H., and Woolf, T. B.

   Networks 2021 2021

   Abs HTML

   The ongoing pandemic has inevitably hastened interest in network epidemiology, resource allocation, and intervention modeling. Perspectives on graph inference and node immunization may be particularly relevant, as municipalities consider appropriate strategies for vaccine distribution, but much of the pertinent literature relies on either complete knowledge of a network’s structure or inference via snapshot. Herein we consider a node immunization task whereby network topology is revealed and immunization decisions are made by agent exploration along a finite-length unbiased random walk. For the purposes of this submission, we focus on scale-free graphs.

2020

1. Accenture

   Markov Decision Processes for Node Immunization with Imperfect Information

   McCabe, L. H.

   Accenture Federal Services 2020

   Abs

   The ongoing COVID-19 pandemic has prompted critical discussion about effective resource allocation strategies, for which perspectives on graph node immunization may inform solutions. One potential limitation, however, is frequent reliance on complete knowledge of a graph’s structure, often derived from anonymized surveillance, demographic data, or assumption. While these approaches provide an essential framework for understanding contagion dynamics and vaccination strategies, complete network information is not always reliable or available in practical situations. We consider the challenge of node immunization where information regarding network topology is realized only through agent exploration along a random walk. We formulate this as a Markov decision process problem and derive heuristic-based policies for scale-free and uncorrelated networks. We demonstrate empirical evidence that these policies achieve their objectives near-optimally and provide a policy-of-policies framework for situations where information about network family does not exist. We also introduce our open-source contagion package and use it to illustrate immunization policy performance with network contagion simulations.