US 12,131,507 B2
Low rank matrix compression
Tomer Bar-On, Petah Tikva (IL); Jacob Subag, Kiryat Haim (IL); Yaniv Fais, Tel Aviv (IL); Jeremie Dreyfuss, Tel-Aviv (IL); Gal Novik, Haifa (IL); Gal Leibovich, Kiryat Yam (IL); Tomer Schwartz, Even Yehuda (IL); Ehud Cohen, Kiryat Motskin (IL); Lev Faivishevsky, Kfar Saba (IL); Uzi Sarel, Zichron-Yaakov (IL); Amitai Armon, Tel-Aviv (IL); and Yahav Shadmiy, Ramat Gan (IL)
Assigned to INTEL CORPORATION, Santa Clara, CA (US)
Filed by Intel Corporation, Santa Clara, CA (US)
Filed on Mar. 28, 2023, as Appl. No. 18/191,565.
Application 18/191,565 is a continuation of application No. 17/344,639, filed on Jun. 10, 2021, granted, now 11,620,766.
Application 17/344,639 is a continuation of application No. 15/482,725, filed on Apr. 8, 2017, granted, now 11,037,330, issued on Jun. 15, 2021.
Prior Publication US 2023/0316589 A1, Oct. 5, 2023
Int. Cl. G06T 9/00 (2006.01); G06N 3/044 (2023.01); G06N 3/045 (2023.01); G06N 3/047 (2023.01); G06N 3/048 (2023.01); G06N 3/084 (2023.01); G06N 3/088 (2023.01); H04N 19/42 (2014.01); H04N 19/436 (2014.01)
CPC G06T 9/002 (2013.01) [G06N 3/044 (2023.01); G06N 3/045 (2023.01); G06N 3/047 (2023.01); G06N 3/048 (2023.01); G06N 3/084 (2013.01); G06N 3/088 (2013.01); H04N 19/42 (2014.11); H04N 19/436 (2014.11)] 12 Claims
OG exemplary drawing
 
1. An apparatus comprising:
processor circuitry coupled to a memory, the processor circuitry to:
facilitate lossy compression to compress data, and to compress one or more weights relating to a neural network layer in a frequency domain, wherein the one or more weights are converted to one or more linearly dependent rows of a matrix, wherein a rank associated with the one or more linearly dependent rows is less than a threshold value as associated with independent rows of the matrix; and
load the matrix into a neural network when the rank is less than the threshold value.