← Back to Sessions

An Agent Based Model to Quantify Gains in Network Processing

Description: 

Explosion monitoring agencies and state-sponsored seismic networks that surveil the globe for signatures of nuclear explosions generally implement some version of what we call an Event Processing Pipeline (EEP) or just a pipeline. Pipelines comprise a sequence of tasks that subject matter experts and their algorithms (collectively called “agents”) apply to detect, locate and characterize the source of such signatures (among other efforts). Both the details of agent implementation, such as whether they work in series or parallel, and their individual success rates, such as the probability that a source depth estimate is accurate, quantify the overall efficacy of an EEP. Researchers and network operators often question if a particular sequence of monitoring functions provides a more complete seismic catalog relative to a reference catalog. Explosion monitoring researchers therefore require the capability to measure how any loss or improvements to agent cooperation and performance impact the likelihood that EEPs will output an accurately characterized special event source, over such a reference (baseline) value. Here, we provide a high-level model of EEP performance and quantify how changes to individual agents improve the overall EEP success rate. In particular, we model an EEP as a series of monitoring function modules that teams of resource-limited agents (working in series or parallel) can achieve with specific probabilities. This model takes the form of a relational object known as a graph. We quantify improvements to EEP over several operational baseline models in terms of such graphs, which directly relate agent performance to that of an EEP. Lastly, we demonstrate our method against a simple example that employs correlation detectors against explosion sourced data.

Session: Network Seismology: Recent Developments, Challenges and Lessons Learned - IV

Type: Oral

Date: 5/2/2024

Presentation Time: 10:30 AM (local time)

Presenting Author: Joshua

Student Presenter: No

Invited Presentation: 

Authors