US 11,922,313 B2
Partitioned machine learning architecture
Bita Darvish Rouhani, San Diego, CA (US); Azalia Mirhoseini, Houston, TX (US); and Farinaz Koushanfar, San Diego, CA (US)
Assigned to WILLIAM MARSH RICE UNIVERSITY, Houston, TX (US)
Appl. No. 16/077,395
Filed by WILLIAM MARSH RICE UNIVERSITY, Houston, TX (US)
PCT Filed Feb. 6, 2017, PCT No. PCT/US2017/016715
§ 371(c)(1), (2) Date Aug. 10, 2018,
PCT Pub. No. WO2017/176356, PCT Pub. Date Oct. 12, 2017.
Claims priority of provisional application 62/294,215, filed on Feb. 11, 2016.
Prior Publication US 2021/0295166 A1, Sep. 23, 2021
Int. Cl. G06N 3/084 (2023.01); G06N 3/045 (2023.01)
CPC G06N 3/084 (2013.01) [G06N 3/045 (2023.01)] 20 Claims
OG exemplary drawing
 
1. A system, comprising:
at least one processor; and
at least one memory including program code which when executed by the at least one processor provides operations comprising:
partitioning, based at least on a resource constraint of a platform, a global machine learning model into a plurality of local machine learning models, each local machine learning model of the plurality of local machine learning models having a subset of a plurality of neurons and interconnections included in the global machine learning model, the global machine learning model being subjected to a depth first partitioning such that each local machine learning model of the plurality of local machine learning models include a same quantity of layers as the global machine learning models, and each local machine learning model of the plurality of local machine learning models having an output layer with a same quantity of neurons as an output layer of the global machine learning model;
transforming training data to at least conform to the resource constraint of the platform; and
training the global machine learning model by at least processing, at the platform, the transformed training data with a first local machine learning model of the plurality of local machine learning models, wherein training the global machine learning model further comprises updating a parameter of the global machine learning model based on at least one corresponding parameter from the plurality of local machine learning models after the transforming of training data.