Is this project an undergraduate, graduate, or faculty project?
Undergraduate
Project Type
individual
Campus
Daytona Beach
Authors' Class Standing
Grace Gratton, Junior
Lead Presenter's Name
Grace Gratton
Lead Presenter's College
DB College of Arts and Sciences
Faculty Mentor Name
Bryan Watson
Abstract
Multi-agent systems are becoming heavily relied upon as the complexity of the world increases. The effectiveness of these systems depends on consensus algorithms; however, the presence of faulted agents can compromise the security and reliability of these consensus algorithms. Therefore, it is crucial to develop robust consensus methods to maintain system security and reliability. Biologically-Inspired Design previously led to the Synchronous Hatching Consensus Algorithm which proved to be robust even with up to 20% of faulted agents reporting false positives. This work aims to provide insights for when the Synchronous Hatching Consensus Algorithm can be applied. This is achieved through three methods: comparing robustness to faulted agents reporting false negatives, performing an uncertainty analysis, and performing a sensitivity analysis. First, an agent-based ANYLOGIC model was tested with 0, 1, 5, 10, 15, and 20 faulted agents reporting false negatives (out of at total population of 100). The model was applied to four separate environments. Robustness to faulted agents was measured by how consistent the hours was to reach 66% consensus across any percentage of faulted agents or environments. A total of 650 iterations were run per faulted agent and environment combination, totaling in 15,600 runs. The model was deemed not robust to faulted agents reporting false negatives. The total probability for a run failing to reach consensus was 59%. The slower changing environments most contributed to the probability a run would fail. The percentage of faulted agents had the second highest impact. The findings indicate that the algorithm should be implemented in an environment which quickly reaches its decision threshold and that when a fault occurs consensus should be assumed, because the model is more robust to false positive faults.
Did this research project receive funding support (Spark, SURF, Research Abroad, Student Internal Grants, Collaborative, Climbing, or Ignite Grants) from the Office of Undergraduate Research?
Yes, Spark Grant
Evaluation of a Biologically-Inspired Multi-Agent System Consensus Algorithm to Develop Application Insights
Multi-agent systems are becoming heavily relied upon as the complexity of the world increases. The effectiveness of these systems depends on consensus algorithms; however, the presence of faulted agents can compromise the security and reliability of these consensus algorithms. Therefore, it is crucial to develop robust consensus methods to maintain system security and reliability. Biologically-Inspired Design previously led to the Synchronous Hatching Consensus Algorithm which proved to be robust even with up to 20% of faulted agents reporting false positives. This work aims to provide insights for when the Synchronous Hatching Consensus Algorithm can be applied. This is achieved through three methods: comparing robustness to faulted agents reporting false negatives, performing an uncertainty analysis, and performing a sensitivity analysis. First, an agent-based ANYLOGIC model was tested with 0, 1, 5, 10, 15, and 20 faulted agents reporting false negatives (out of at total population of 100). The model was applied to four separate environments. Robustness to faulted agents was measured by how consistent the hours was to reach 66% consensus across any percentage of faulted agents or environments. A total of 650 iterations were run per faulted agent and environment combination, totaling in 15,600 runs. The model was deemed not robust to faulted agents reporting false negatives. The total probability for a run failing to reach consensus was 59%. The slower changing environments most contributed to the probability a run would fail. The percentage of faulted agents had the second highest impact. The findings indicate that the algorithm should be implemented in an environment which quickly reaches its decision threshold and that when a fault occurs consensus should be assumed, because the model is more robust to false positive faults.