US 11,741,381 B2
Weighted adaptive filtering based loss function to predict the first occurrence of multiple events in a single shot
V Kishore Ayyadevara, Hyderabad (IN); Sree Harsha Ankem, Hyderabad (IN); Raghav Bali, Bangalore (IN); Rohan Khilnani, Hyderabad (IN); Vineet Shukla, Bangalore (IN); Saikumar Chintareddy, Hyderabad (IN); and Ranraj Rana Singh, Thane (IN)
Assigned to OPTUM TECHNOLOGY, INC., Eden Prairie, MN (US)
Filed by Optum Technology, Inc., Eden Prairie, MN (US)
Filed on Jul. 14, 2020, as Appl. No. 16/928,616.
Prior Publication US 2022/0019913 A1, Jan. 20, 2022
Int. Cl. G06N 5/04 (2023.01); G06N 20/00 (2019.01)
CPC G06N 5/04 (2013.01) [G06N 20/00 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
identifying, by one or more processors, a plurality of training data fields, wherein a training data field of the plurality of training data fields is associated with a training input set and an event label set;
determining, by the one or more processors, an inferred event probability set for the training data field of the plurality of training data fields using a multi-event-type prediction model and based at least in part on the training input set associated with the training data field,
wherein (i) the multi-event-type prediction model is trained based at least in part on a cross-event type loss value that is determined based at least in part on a per-event-type loss value for an event type of a plurality of event types, (ii) the per-event-type loss value for the event type is determined based at least in part on (a) an event label set for the training data field in one or more related subsets of the plurality of training data fields for the event type and (b) the inferred event probability set for the training data field in the one or more related subsets, and (iii) a related subset of the one or more related subsets of the plurality of training data fields for the event type comprises those training data fields of the plurality of training data fields having an event label set that indicates the training data field describes no-occurrence of the event type or first-occurrence of the event type in relation to the training data field;
generating, by the one or more processors, a first-occurrence prediction based at least in part on the multi-event-type prediction model, wherein the first-occurrence prediction comprises a first-occurrence prediction item for the event type of the plurality of event types; and
initiating, by the one or more processors, the performance of one or more prediction-based actions based at least in part on the first-occurrence prediction.