US 11,700,014 B2
Weight data compression method, weight data decompression method, weight data compression device, and weight data decompression device
Yoshinobu Hashimoto, Yokohama (JP)
Assigned to SOCIONEXT INC., Kanagawa (JP)
Filed by Socionext Inc., Kanagawa (JP)
Filed on Apr. 6, 2022, as Appl. No. 17/714,631.
Claims priority of application No. 2021-068040 (JP), filed on Apr. 13, 2021.
Prior Publication US 2022/0329259 A1, Oct. 13, 2022
Int. Cl. H03M 5/00 (2006.01); H03M 7/30 (2006.01); G06N 3/08 (2023.01); H03M 7/46 (2006.01)
CPC H03M 7/6011 (2013.01) [G06N 3/08 (2013.01); H03M 7/6005 (2013.01); H03M 7/70 (2013.01); H03M 7/46 (2013.01)] 14 Claims
OG exemplary drawing
 
1. A weight data compression method of compressing ternary weight data used in a neural network, the weight data compression method comprising:
generating a 4-bit data string by dividing the ternary weight data into data items each having 4 bits, the ternary weight data being a data string of data items each having any one of three 2-bit values, the 4-bit data string being a data string of 4-bit data items each expressed as any one of nine 4-bit values; and
generating first compressed data including a first flag value string and a first non-zero value string by generating the first flag value string and the first non-zero value string, the first flag value string being generated by assigning one of 0 and 1 as a first flag value of a 1-bit flag to a 4-bit data item 0000 and assigning an other of 0 and 1 as a second flag of the 1-bit flag to a 4-bit data item other than 0000 among the 4-bit data items in the 4-bit data string, the first non-zero value string being generated by converting the 4-bit data item other than 0000 into a 3-bit data item having any one of eight 3-bit values.