US 12,008,075 B2
Training federated learning models
Shoichiro Watanabe, Tokyo (JP); Kenichi Takasaki, Tokyo (JP); Mari Abe Fukuda, Tokyo (JP); Sanehiro Furuichi, Tokyo (JP); and Yasutaka Nishimura, Yamato (JP)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Aug. 16, 2021, as Appl. No. 17/402,764.
Prior Publication US 2023/0050708 A1, Feb. 16, 2023
Int. Cl. G06F 18/214 (2023.01); G06F 17/18 (2006.01); G06F 18/21 (2023.01); G06N 3/08 (2023.01)
CPC G06F 18/2148 (2023.01) [G06F 17/18 (2013.01); G06F 18/217 (2023.01); G06N 3/08 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for training a federated learning model, the method comprising:
distributing a federated learning model to a plurality of computing nodes, wherein each computing node includes a set of local training data for training the federated learning model comprising data samples each labeled with one label of a plurality of labels;
receiving statistical data from each of the plurality of computing nodes, wherein the statistical data for a computing node is based on the local training data of the computing node, and wherein the statistical data indicates a count of data samples for each label;
analyzing the statistical data to identify one or more computing nodes having local training data in which a ratio of a count of data samples for one or more labels to a count of data samples for a most-represented label in the local training data is below a threshold value;
providing additional training data to the identified one or more computing nodes, wherein the additional training data comprises data samples labeled with labels other than the most-represented label, and wherein the identified one or more computing nodes train the federated learning model using the set of local training data and the additional training data; and
receiving results from the plurality of computing nodes in response to the plurality of computing nodes each locally training the federated learning model, and generating a trained global model based on the results.