US 11,740,905 B1
Drift detection in static processes
Rajaram Kudli, Bengaluru (IN); Satish Padmanabhan, Sunnyvale, CA (US); Fuk Ho Pius Ng, Portland, OR (US); and Dushyanth Gokhale, Bengaluru (IN)
Assigned to DIMAAG-AI, Inc., Fremont, CA (US)
Filed by DIMAAG-AI, Inc., Fremont, CA (US)
Filed on Jul. 25, 2022, as Appl. No. 17/814,684.
Int. Cl. G06N 5/022 (2023.01); G06F 9/30 (2018.01)
CPC G06F 9/30192 (2013.01) [G06N 5/022 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A method comprising:
determining via one or more processors a drift value by comparing a first plurality of feature values corresponding with a first time period with a second plurality of feature values corresponding with a second time period;
determining via one or more processors a predicted outcome value by applying a prediction model to a third plurality of feature values;
determining via one or more processors a plurality of sensitivity values corresponding with the third plurality of feature values when it is determined that the drift value crosses a designated drift threshold, a designated sensitivity value of the plurality of sensitivity values indicating a degree to which change in a designated feature value affects the predicted outcome value, the designated sensitivity value being determined by re-applying the prediction model while varying the designated feature value and holding fixed other feature values, wherein the third plurality of feature values correspond with a plurality of features determined by a data generating process;
selecting a designated subset of the third plurality of feature values based on the plurality of sensitivity values, the designated subset of the third plurality of feature values corresponding with a subset of the plurality of features; and
transmitting an instruction via a communication interface to update one or more control parameters in the data generating process corresponding with the designated subset of the third plurality of features based on the plurality of sensitivity values.