The Defense Advanced Research Projects Agency (DARPA) Defense Sciences Office (DSO) is requesting information on methods for representation and extraction of knowledge in scientific data analysis and modeling to support computational reasoning over diverse sources of evidence. Scientific discoveries rely on piecing together evidence from many sources-including mathematical, computational, experimental, and observational-spanning multiple modalities and scales.  Because of the diversity of the data types and reasoning involved, this fusion generally only occurs in the brains of researchers, resulting in fragmented understanding and leading to a significant bottleneck to new discoveries.  New representations and algorithms for connecting such heterogeneous forms of scientific information would greatly accelerate the research process.

The DARPA/DSO Simplifying Complexity in Scientific Discovery (SIMPLEX) program has developed a variety of tools and methods to aid scientific research exploiting multimodal data.  While SIMPLEX has led to new capabilities in a number of domains, DSO is looking to extend and conceptually unify these results, which span both general purpose abstract methods and domain-specific application tools.  The apparent tension between domain-specific tools requiring tuning to specific problems and general purpose abstract methods that require idealization of the data stems from an underlying lack of a properly rigorous mathematical framework for automatic knowledge extraction, which is the focus of this RFI.  

See DARPA-SN-17-32.