US 12,456,059 B2
Dynamically parameterized machine learning frameworks
Irfan Bulu, Sartell, MN (US); and Gregory D. Lyng, Minneapolis, MN (US)
Assigned to UnitedHealth Group Incorporated, Minnetonka, MN (US)
Filed by UnitedHealth Group Incorporated, Minnetonka, MN (US)
Filed on Sep. 24, 2021, as Appl. No. 17/484,571.
Prior Publication US 2023/0097965 A1, Mar. 30, 2023
Int. Cl. G06N 5/02 (2023.01); G06N 3/004 (2023.01); G06N 5/04 (2023.01)
CPC G06N 5/027 (2013.01) [G06N 3/004 (2013.01); G06N 5/04 (2013.01)] 17 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
generating, by one or more processors and using a target model of a dynamically parameterized machine learning framework, a predictive output for a predictive input, wherein the dynamically parameterized machine learning framework comprises (i) an encoder machine learning model and a decoder machine learning model that comprise a plurality of statistically generated parameters and (ii) the target model, which comprises an initial plurality of dynamically generated parameters, wherein (a) the plurality of statistically generated parameters is fixed after training of the dynamically parameterized machine learning framework, (b) the initial plurality of dynamically generated parameters is updated during a plurality of processing iterations of the dynamically parameterized machine learning framework, (c) the plurality of processing iterations comprises a training processing iteration and an inferential processing iteration, and (d) and wherein generating the predictive output comprises:
determining, using the encoder machine learning model, one or more dynamically generated parameters for the target model based at least in part on the predictive input,
updating the initial plurality of dynamically generated parameters based on the one or more dynamically generated parameters without updating the plurality of statistically generated parameters,
determining, using the target model, and based at least in part on the one or more dynamically generated parameters, a target model output for the predictive input, and
determining the predictive output based at least in part on the target model output; and
initiating, by the one or more processors, one or more prediction-based actions based at least in part on the predictive output.