Case Study: U.S. Navy reduces ML model update time by 97% with Fiddler

U.S. Navy PMS 408 Reduces ML Model Update Time by 97% with Help From Fiddler

The U.S. Navy’s NAVSEA Expeditionary Missions Program Office (PMS 408) needed a better way to monitor and update its computer vision-based Automatic Target Recognition (ATR) models used for underwater explosive ordinance disposal. Working with Fiddler and the Defense Innovation Unit, the Navy sought a more efficient MLOps approach to keep models accurate as threats and data changed over time.

Fiddler helped build the Automated Machine Learning for Mine Countermeasures Operations (AMMO) prototype, adding ML monitoring, explainability, and visual debugging to surface model issues and support faster root-cause analysis. As a result, the Navy reduced model update time by 97%, improved transparency into model decisions, and moved the prototype into production with the Naval Information Warfare Center Pacific.

U.S. Navy