Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Geospatial Trajectory Generation via Efficient Abduction: Deployment for Independent Testing

Divyagna Bavikadi, Dyuman Aditya, Devendra Parkar, Paulo Shakarian, Graham Mueller, Chad Parvis, Gerardo I. Simari

TL;DR

This work tackles the generation of realistic yet synthetic human trajectories within geospatial constraints by framing it as abduction guided by a parsimony function defined as an aggregate truth value over an annotated temporal logic program. The main method combines a provable lower-bound bound on parsimony via subsets of the logic program with an informed search using $A^*$, alongside a scalable software stack deployed on AWS that integrates knowledge graphs and rule learning. Key contributions include a formal bound on parsimony, a single-hop heuristic that accelerates search, and empirical validation showing exact, scalable trajectory generation that often evades ML anomaly detectors, plus independent government testing demonstrating practical deployment. This approach enhances explainability through rule-based deductions and supports privacy-preserving testing by producing plausible synthetic trajectories for security research with real-world impact in government evaluation contexts.

Abstract

The ability to generate artificial human movement patterns while meeting location and time constraints is an important problem in the security community, particularly as it enables the study of the analog problem of detecting such patterns while maintaining privacy. We frame this problem as an instance of abduction guided by a novel parsimony function represented as an aggregate truth value over an annotated logic program. This approach has the added benefit of affording explainability to an analyst user. By showing that any subset of such a program can provide a lower bound on this parsimony requirement, we are able to abduce movement trajectories efficiently through an informed (i.e., A*) search. We describe how our implementation was enhanced with the application of multiple techniques in order to be scaled and integrated with a cloud-based software stack that included bottom-up rule learning, geolocated knowledge graph retrieval/management, and interfaces with government systems for independently conducted government-run tests for which we provide results. We also report on our own experiments showing that we not only provide exact results but also scale to very large scenarios and provide realistic agent trajectories that can go undetected by machine learning anomaly detectors.

Geospatial Trajectory Generation via Efficient Abduction: Deployment for Independent Testing

TL;DR

This work tackles the generation of realistic yet synthetic human trajectories within geospatial constraints by framing it as abduction guided by a parsimony function defined as an aggregate truth value over an annotated temporal logic program. The main method combines a provable lower-bound bound on parsimony via subsets of the logic program with an informed search using , alongside a scalable software stack deployed on AWS that integrates knowledge graphs and rule learning. Key contributions include a formal bound on parsimony, a single-hop heuristic that accelerates search, and empirical validation showing exact, scalable trajectory generation that often evades ML anomaly detectors, plus independent government testing demonstrating practical deployment. This approach enhances explainability through rule-based deductions and supports privacy-preserving testing by producing plausible synthetic trajectories for security research with real-world impact in government evaluation contexts.

Abstract

The ability to generate artificial human movement patterns while meeting location and time constraints is an important problem in the security community, particularly as it enables the study of the analog problem of detecting such patterns while maintaining privacy. We frame this problem as an instance of abduction guided by a novel parsimony function represented as an aggregate truth value over an annotated logic program. This approach has the added benefit of affording explainability to an analyst user. By showing that any subset of such a program can provide a lower bound on this parsimony requirement, we are able to abduce movement trajectories efficiently through an informed (i.e., A*) search. We describe how our implementation was enhanced with the application of multiple techniques in order to be scaled and integrated with a cloud-based software stack that included bottom-up rule learning, geolocated knowledge graph retrieval/management, and interfaces with government systems for independently conducted government-run tests for which we provide results. We also report on our own experiments showing that we not only provide exact results but also scale to very large scenarios and provide realistic agent trajectories that can go undetected by machine learning anomaly detectors.
Paper Structure (10 sections, 1 theorem, 1 equation, 6 figures, 1 table)

This paper contains 10 sections, 1 theorem, 1 equation, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Motivating Application
  3. Technical Preliminaries
  4. Efficient and Scalable Geospatial Abduction for Trajectory Generation
  5. Software Stack
  6. Internal Evaluation
  7. Deployment for Independent Evaluation
  8. Related Work
  9. Conclusion
  10. Acknowledgement

Key Result

Theorem 4.1

For ground atom $b$, timepoint $t$, $\Pi' \subseteq \Pi$, and $I' \preceq I$, we have $\Gamma^\ast_{\Pi'}(I') (b, t) \sqsubseteq \Gamma^\ast_{\Pi}(I) (b, t)$.

Figures (6)

  • Figure 1: Left: The graph represents the city of Knoxville with landmarks, and the line plot denotes part of an agent's sub-trajectory. Right: Generated trajectory.
  • Figure 2: Visual representation of deployment.
  • Figure 3: Left: Value with subset logic program, logic program $\Pi', \Pi$ for the Knoxville location. Right: Runtime Comparison with Depth-First Search and A* Search.
  • Figure 4: Left: Relative Anomaly Ratio. Anomalies in generated movements when compared with anomalies found in the training data by an ensemble ML anomaly detection algorithm. Right: Runtime speedup due to ad-hoc weighting.
  • Figure 5: Visual depiction of rules from Table \ref{['tab:examplerlsRules']}. Left: The top rule. Right: Both rules.
  • ...and 1 more figures

Theorems & Definitions (3)

  • Example 3.1: Language and Syntax
  • Example 3.2
  • Theorem 4.1