US 11,676,068 B1
Method, product, and apparatus for a machine learning process leveraging input sparsity on a pixel by pixel basis
Michael Patrick Zimmer, Chicago, IL (US); Ngai Ngai William Hung, San Jose, CA (US); Yong Liu, Cupertino, CA (US); and Dhiraj Goswami, Wilsonville, OR (US)
Assigned to Cadence Design Systems, Inc., San Jose, CA (US)
Filed by Cadence Design Systems, Inc., San Jose, CA (US)
Filed on Jun. 30, 2020, as Appl. No. 16/946,673.
Int. Cl. G06N 20/00 (2019.01); G06F 17/16 (2006.01)
CPC G06N 20/00 (2019.01) [G06F 17/16 (2013.01)] 27 Claims
OG exemplary drawing
 
1. A machine learning apparatus, comprising:
a data memory for storing a data matrix;
an array input controller for controlling execution of a machine learning processing job on a pixel by pixel basis; and
a systolic array for executing the machine learning processing job, wherein the array input controller
determines a maximum number data values for input into each row of a plurality of rows of the systolic array based on a count of a number of non-zero data values in data for input into each row and
selects up to the maximum number of non-zero data values for input into each row of the plurality of rows of the systolic array.