US 12,353,963 B2
Selective bit inversion in storage operations for machine learning
Daren Croxford, Swaffham Prior (GB); Sharjeel Saeed, Cambridge (GB); Rachel Jean Trimble, Grindleford (GB); and Timothy Fawcett Milner, Cambridge (GB)
Assigned to Arm Limited, Cambridge (GB)
Filed by Arm Limited, Cambridge (GB)
Filed on Jan. 11, 2021, as Appl. No. 17/145,804.
Prior Publication US 2022/0222569 A1, Jul. 14, 2022
Int. Cl. G06N 20/00 (2019.01); G06F 3/06 (2006.01)
CPC G06N 20/00 (2019.01) [G06F 3/0604 (2013.01); G06F 3/0653 (2013.01); G06F 3/0673 (2013.01)] 17 Claims
OG exemplary drawing
 
1. A processing unit comprising:
volatile storage for storing machine learning data, wherein the volatile storage is configured to store machine learning data in binary representation;
a data processing engine communicatively coupled to the volatile storage and for performing machine learning operations using the machine learning data,
wherein the processing unit is configured to selectively invert bit values in binary representations of portions of the machine learning data when performing storage operations using the volatile storage by:
processing the machine learning data to obtain a selected portion of the machine learning data and one or more unselected portions of the machine learning data, wherein the selected portion is selected based on at least one characteristic of the selected portion of the machine learning data; and
selectively inverting bit values by inverting at least one bit value in a binary representation of the selected portion of machine learning data and without inverting bit values in binary representations of the unselected portions of the machine learning data,
wherein the storage operations comprise at least one of:
writing the selected and unselected portions of the machine learning data to the volatile storage; and
reading the selected and unselected portions of the machine learning data from the volatile storage.