Charles River Analytics developed State Estimation via Asynchronous Probabilistic Inference for Logistics Enterprises (STAPLES) under a $2 million contract awarded by the Defense Advanced Research Projects Agency (DARPA). STAPLES is an instrumental part of DARPA’s LogX program, which enhances how the DoD’s Joint Logistics Enterprise operates in an increasingly contested global security environment. The Joint Logistics Enterprise spans supply chain and logistics operations, ensuring forces are ready to go at a moment’s notice. LogX will improve the enterprise’s ability for real-time situational awareness, future state prediction, and assessment.
STAPLES software offers highly performant, distributed, dynamic state estimation and prediction at scale using asynchronous belief propagation, a tractable-by-construction (TBC) approach, and a temporal querying interface. It is implemented within Scruff™, the latest probabilistic programming framework from Charles River Analytics.
Performant, accurate, and automatic inference across a wide range of models is a “holy grail” of probabilistic programming—many developers can write high quality probabilistic programs with little education, but understanding the tradeoffs and challenges of tractable inference algorithm development and application requires significantly more expertise. Scruff, unlike other probabilistic programming frameworks, is designed to enable high performance inference with minimal expertise in inference algorithms.
Scruff is a next-generation probabilistic programming framework built on the success of the Figaro™ probabilistic programming language, which has been applied successfully to solve a diverse range of DoD problems.