Confidence Estimation

Confidence Estimation

Understanding uncertainty in a measurement and models is critical to making informed decisions. At GDA we provide tools to estimate and track the uncertainty of individual sensors, sensor network coverage, and mathematical models to support informed decisions based on understanding probability of correctness for a predicted result.

Field Geometric Health - Sensor Coverage

Sensor networks work together to detect events and track objects of interest as they move through an environment. Understanding the coverage of the network and the probability of an event being detected is critical for risk assessment and planning. GDA uses topographic data analysis techniques to estimate both the coverage and accuracy of system measurements to provide both situational awareness and manage planning under uncertainty.

Field Geometric Health of hydrophone sensors under different wave conditions as a function of sea state on the Beaufort Scale. In this case, density as indicates probability of event detection. (DARPA Ocean of Things Phase I 2020)

Field Geometric Health - Communication Coverage

Having access to a reliable network connection is necessary to keep people in the field up to date with conditions in a rapidly changing environment. This is particularly important in wildfire response where it is critical that firefighters ground have the information available to operate efficiently and safely. To help address this concern, GDA applies Field Geometric Health algorithms to estimate predicted bandwidth availability at specific locations and identifying areas where adding new hot spots would have the greatest impact.

The Field Geometric Health of a set of networked hotspots providing communication coverage as part of a simulated wildfire response. (DARPA Phase II SBIR - 2023)

Verification and Validation

GDA performs fundamental research in the verification and validation (V&V) of AI, with the goal of providing certificates of novelty robustness for specific AI systems. Based on our expertise in persistent homology, our V&V toolset identifies blind spots and corner cases in the data, feature, and decision spaces of an AI system, giving users a proactive approach to catastrophe mitigation