US 11,842,256 B2
Ensemble training in a distributed marketplace
Killian Levacher, Dundrum (IE); Emanuele Ragnoli, Mulhuddart (IE); Stefano Braghin, Dublin (IE); and Gokhan Sagirlar, Dublin (IE)
Assigned to International Business Machines Corporation Armonk, New York, NY (US)
Filed by INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed on May 15, 2020, as Appl. No. 15/929,683.
Prior Publication US 2021/0357819 A1, Nov. 18, 2021
Int. Cl. G06N 20/20 (2019.01); H04L 9/00 (2022.01); G06N 3/088 (2023.01)
CPC G06N 20/20 (2019.01) [G06N 3/088 (2013.01); H04L 9/50 (2022.05)] 20 Claims
OG exemplary drawing
 
1. A method for ensemble training of machine learning models using one or more processors, comprising:
receiving, by a blockchain, a machine learning model request and a dataset split into training data and ensemble training data, wherein the machine learning model request specifies a selected performance threshold to perform a function on the dataset;
generating and recording a hash of the training data, the ensemble training data, and the selected performance threshold on the blockchain;
training, by each of a plurality of machine learning nodes, a respective machine learning model to aggregately generate a plurality of machine learning models by the plurality of machine learning nodes, each respective machine learning model trained under a selected one of differing model architectures, wherein each of the plurality of machine learning models trained by the plurality of machine learning nodes competes with one another to perform the function most optimally with respect to performance, and wherein the blockchain is used to incentivize each of the plurality of machine learning nodes to perform the training of the respective machine learning model utilizing common data to identify one or more of the plurality of machine learning models which most optimally perform the function with respect to the performance;
generating and recording a hash of model data of each of the plurality of machine learning models on the blockchain;
analyzing each of the plurality of machine learning models to identify weights applied to each respective machine learning model, and combining the plurality of machine learning models into one or more ensemble machine learning models according to the weights;
generating and recording a hash of data associated with the weights and the one or more ensemble machine learning models on the blockchain; and
providing the one or more ensemble machine learning models that perform the function on the dataset at the performance of equal or greater to the selected performance threshold.