IARPA wants to know how it can recognize and correct for ‘model drift’ in its research models
The Intelligence Advanced Research is Projects Activity (IARPA) is looking for some clear and imaginative thinking that could help the agency recognize when some of the intellectual “models” that underpin its research are beginning to suffer from what it calls “model drift,” which can make the research far less useful.
IARPA, says a notice it published on January 28, “is seeking information on automated methods to detect, quantify, and correct model drift.”
The Intelligence Community needs to know when its algorithms require re-training, when assumptions underlying a model no longer apply, when the input data are significantly different from the data on which the model was trained, when the context in which events are being forecasted has changed significantly, and when confidence in the output of a model has diminished enough to require re-training.
IARPA has issued a Request for Information (RFI) which seeks input and ideas from interested prospective vendors and researchers. These responses, which should consist of a maximum of five pages, can be submitted to IARPA by March 14, says the agency’s notice.
IARPA would like to know what are the automated methods for detecting, quantifying and correcting model drift.
“IARPA requests that respondents submit ideas related to this topic for use by the Government in formulating a potential program,” the notice continues.
Further information is available from Dewey Murdick, of IARPA, at firstname.lastname@example.org.