Methods for compressing and decompressing weighted data performed in the processor of a computer device

Abstract

The invention relates to a method for compressing and decompressing weighted data performed in a computer processing unit. A possible scheme might include a step of compressing weighted data by obtaining a set of weighted data of binary numbers representing the weighted values, creating a frame payload consisting of a compressed first frame of the first subset of the weighted values ​​in the weighted data set, and creating a compressed weighted data block containing the frame payload. One possible approach might include a step to decompress the weighted data by retrieving the compressed weighted data block, where the compressed weighted data block includes the frame header associated with the frame payload, where the frame header includes the normalization coefficient indicator and where the frame payload includes the compressed weight values, and creating the first decompressed frame which includes the decompressed weight values ​​of the compressed weight values ​​of the frame payload.