US 11,790,226 B2
Model compression by sparsity—inducing regularization optimization
Tianyi Chen, Redmond, WA (US); Sheng Yi, Redmond, WA (US); Yixin Shi, Redmond, WA (US); and Xiao Tu, Medina, WA (US)
Assigned to MICROSOFT TECHNOLOGY LICENSING, LLC, Redmond, WA (US)
Filed by MICROSOFT TECHNOLOGY LICENSING, LLC, Redmond, WA (US)
Filed on Jun. 1, 2020, as Appl. No. 16/889,775.
Prior Publication US 2021/0390384 A1, Dec. 16, 2021
Int. Cl. G06N 3/08 (2023.01); G06N 3/04 (2023.01)
CPC G06N 3/08 (2013.01) [G06N 3/04 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method of compressing a machine learning model, the method comprising:
receiving a machine learning model implemented in a computing system;
training the machine learning model using a first set of training data;
executing a sparsity-inducing regularization optimization process on the machine learning model, wherein execution of the sparsity-inducing regularization optimization process comprises using switching parameters to control a period for an orthant step and a proximal gradient method;
based on the sparsity-inducing regularization optimization process, receiving a compressed machine learning model; and
executing the compressed machine learning model to generate one or more outputs.