Method and apparatus with data processing
AssigneesSamsung Electronics Co., Ltd., Seoul National University R&Db Foundation
InventorsYuhwan Ro, Byeongho Kim, Jaehyun Park, Jungho Ahn, Minbok Wi, Sunjung Lee, Eojin Lee, Wonkyung Jung, Jongwook Chung, Jaewan Choi
Life of this patent
Estimating the term…
Abstract
A processor-implemented data processing method includes: generating compressed data of first matrix data based on information of a distance between valid elements included in the first matrix data; fetching second matrix data based on the compressed data; and generating output matrix data based on the compressed data and the second matrix data.
Claims19
A processor-implemented data processing method, the method comprising:
- generating compressed data of first matrix data of a neural network based on a sparsity characteristic of the first matrix data;
- partitioning the compressed data into rows;
- respectively storing pieces of the partitioned compressed data in first regions of a plurality of banks;
- copying portions of the pieces of the partitioned compressed data stored in the first regions and storing the portions in a second region of one of the plurality of banks.
A data processing apparatus comprising:
- a memory;
- and
- one or more processors configured to:
- generate compressed data of first matrix data of a neural network based on a sparsity characteristic of the first matrix data;
- partition the compressed data into rows;
- respectively store pieces of the partitioned compressed data in first regions of a plurality of banks of the memory;
- copy portions of the pieces of the partitioned compressed data stored in the first regions and storing the portions in a second region of one of the plurality of banks of the memory.
Dependencies are inferred from the claim text.