US 11,669,724 B2
Machine learning using informed pseudolabels
Philip A. Sallee, South Riding, VA (US); James Mullen, Leesburg, VA (US); and Franklin Tanner, Ashburn, VA (US)
Assigned to Raytheon Company, Waltham, MA (US)
Filed by Raytheon Company, Waltham, MA (US)
Filed on May 16, 2019, as Appl. No. 16/413,730.
Claims priority of provisional application 62/810,113, filed on Feb. 25, 2019.
Claims priority of provisional application 62/672,758, filed on May 17, 2018.
Prior Publication US 2019/0354857 A1, Nov. 21, 2019
Int. Cl. G06N 3/08 (2023.01); G06N 3/04 (2023.01)
CPC G06N 3/08 (2013.01) [G06N 3/04 (2013.01)] 10 Claims
OG exemplary drawing
 
1. A method of training a machine learning (ML) model using informed pseudolabels, the method performed by circuitry configured to implement a pseudolabel generator, the method comprising:
receiving previously assigned labels indicating an expected classification for data, the labels generated by a partially trained ML model in an immediately previous training epoch and the labels having a specified uncertainty;
generating the informed pseudolabels for the data based on a mathematical combination of (i) a first probability of the previously assigned labels given a class vector determined by the partially trained ML model, and (ii) a second probability of the class vector determined by the partially trained ML model given the data; and
training, by substituting the informed pseudolabels for the previously assigned labels in a next epoch of training, the partially trained ML model resulting in the ML model.