US 11,914,670 B2
Methods and systems for product quantization-based compression of a matrix
Krtin Kumar, Montreal (CA); Mehdi Rezagholizadeh, Montreal (CA); and Peyman Passban, Montreal (CA)
Assigned to HUAWEI TECHNOLOGIES CO., LTD., Shenzhen (CN)
Filed by Krtin Kumar, Montreal (CA); Mehdi Rezagholizadeh, Montreal (CA); and Peyman Passban, Montreal (CA)
Filed on Sep. 8, 2020, as Appl. No. 17/014,712.
Prior Publication US 2022/0075843 A1, Mar. 10, 2022
Int. Cl. G06F 17/16 (2006.01); G06F 40/284 (2020.01); H03M 7/30 (2006.01); G06N 3/08 (2023.01)
CPC G06F 17/16 (2013.01) [G06F 40/284 (2020.01); G06N 3/08 (2013.01); H03M 7/70 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computing system comprising:
a memory; and
a processing device in communication with the memory, the processing device configured to execute instructions to cause the computing system to compress a matrix including:
partition the matrix, having a plurality of rows formed by a respective plurality of vectors, into a plurality of submatrices, each submatrix containing sub-vectors from a respective group of one or more contiguous columns of the matrix;
for each given submatrix, cluster the sub-vectors of the given submatrix into a plurality of clusters;
for each given cluster, compute and store, in the memory, a centroid and a variance based on the sub-vectors belonging to the given cluster; and
store, in the memory, a mapping relating each vector to a respective cluster in each submatrix;
wherein the stored centroids, stored variances and stored mapping form a set of compressed data for reconstruction of the matrix, the set of compressed data requiring fewer bits to store in the memory than the matrix prior to compression.