Compression of high dynamic ratio fields for machine learning

Patent metadata

Field	Value
Publication number	US-11025271-B2
Application number	US-202016798186-A
Country	US
Kind code	B2
Filing date	Feb 21, 2020
Priority date	Feb 22, 2019
Publication date	Jun 1, 2021
Grant date	Jun 1, 2021

Abstract

Various embodiments include methods and devices for implementing compression of high dynamic ratio fields. Various embodiments may include receiving a compression block having data units, receiving a mapping for the compression block, wherein the mapping is configured to map bits of each data unit to two or more data fields to generate a first set of data fields and a second set of data fields, compressing the first set of data fields together to generate a compressed first set of data fields, and compressing the second set of data fields together to generate a compressed second set of data fields.

First claim

What is claimed is:

1. A method of compressing data, comprising:
receiving, by a processor, a compression block having data units;
receiving, by the processor, a mapping for the compression block, wherein the mapping is configured to map bits of each data unit to two or more data fields to generate a first set of data fields and a second set of data fields;
compressing, by the processor, the first set of data fields together to generate a compressed first set of data fields; and
compressing, by the processor, the second set of data fields together to generate a compressed second set of data fields.

2. The method of claim 1 , wherein the mapping is further configured to map the bits of each data unit to the two or more data fields based on any of sizes of the data units, types of the data units, locality of portions of the data units, or estimated compression ratios of portions of the data units.

3. The method of claim 1 , further comprising separating the bits of each data unit into the two or more data fields according to the mapping.

4. The method of claim 1 , further comprising:
analyzing a plurality of mappings for the compression block for bits of the data units mapped to data fields having a locality exceeding a locality threshold;
estimating a compression ratio for mappings having locality exceeding the locality threshold; and
selecting the mapping having the highest estimated compression ratio.

5. The method of claim 1 , wherein:
compressing the first set of data fields together to generate a compressed first set of data fields comprises compressing the first set of data fields using a first compression method;
compressing the second set of data fields together to generate a compressed second set of data fields comprises compressing the second set of data fields using a second compression method; and
the first compression method is different from the second compression method.

6. The method of claim 1 , further comprising dynamically generating a mapping for the compression block, including:
executing a machine learning algorithm with data of a reconstructed compression block;
updating a mapping machine learning algorithm with compression statistics resulting from execution of the machine learning algorithm with the data of the reconstructed compression block; and
executing the mapping machine learning algorithm to generate mapping parameters for compressing the compression block.

7. The method of claim 1 , further comprising dynamically generating a mapping for the compression block, including:
executing a machine learning algorithm with data of a reconstructed compression block;
associating compression statistics resulting from execution of the machine learning algorithm with a compression ratio of the compression block; and
executing a directed search engine to generate mapping parameters for compressing the compression block.

8. The method of claim 1 , wherein:
the data units each have a size of M bits that is greater than a quantum of size N bits for compression; and
the data fields each have a size no greater than N bits.

9. A computing device, comprising:
a processing device configured with processing device-executable instructions to cause the processing device to execute operations comprising:
receiving a compression block having data units;
receiving a mapping for the compression block, wherein the mapping is configured to map bits of each data unit to two or more data fields to generate a first set of data fields and a second set of data fields; and

a compression engine configured to:
compress the first set of data fields together to generate a compressed first set of data fields; and
compress the second set of data fields together to generate a compressed second set of data fields.

10. The computing device of claim 9 , wherein the mapping is further configured to map the bits of each data unit to the two or more data fields based on any of sizes of the data units, types of the data units, locality of portions of the data units, or estimated compression ratios of portions of the data units.

11. The computing device of claim 9 , wherein the processing device is configured with processing device-executable instructions to perform operations further comprising separating the bits of each data unit into the two or more data fields according to the mapping.

12. The computing device of claim 11 , wherein the processing device is configured with processing device-executable instructions to perform operations further comprising:
analyzing a plurality of mappings for the compression block for bits of the data units mapped to data fields having a locality exceeding a locality threshold;
estimating a compression ratio for mappings having locality exceeding the locality threshold; and
selecting the mapping having the highest estimated compression ratio.

13. The computing device of claim 9 , wherein the processing device is configured with processing device-executable instructions to perform operations such that:
compressing the first set of data fields together to generate a compressed first set of data fields comprises compressing the first set of data fields using a first compression method;
compressing the second set of data fields together to generate a compressed second set of data fields comprises compressing the second set of data fields using a second compression method; and
the first compression method is different from the second compression method.

14. The computing device of claim 9 , wherein the processing device is configured with processing device-executable instructions to perform operations further comprising dynamically generating a mapping for the compression block, including:
executing a machine learning algorithm with data of a reconstructed compression block;
updating a mapping machine learning algorithm with compression statistics resulting from execution of the machine learning algorithm with the data of the reconstructed compression block; and
executing the mapping machine learning algorithm to generate mapping parameters for compressing the compression block.

15. The computing device of claim 9 , wherein the processing device is configured with processing device-executable instructions to perform operations further comprising dynamically generating a mapping for the compression block, including:
executing a machine learning algorithm with data of a reconstructed compression block;
associating compression statistics resulting from execution of the machine learning algorithm with a compression ratio of the compression block; and
executing a directed search engine to generate mapping parameters for compressing the compression block.

16. The computing device of claim 9 , wherein:
the data units each have a size of M bits that is greater than a quantum of size N bits for compression for a compression block; and
the data fields each have a size no greater than N bits.

17. A computing device, comprising:
means for receiving a compression block having data units;
means for receiving a mapping for the compression block, wherein the mapping is configured to map bits of each data unit to two or more data fields to generate a first set of data fields and a set plurality of data fields;
means for compressing the first set of data fields together to generate a compressed first set of data fields; and
means for compressing the second set of data fields together to generate a compressed second set of data fields.

18. The computing device of claim 17 , wherein the mapping i

Assignees

• Qualcomm Inc

Inventors

• Sung Clara Ka Wah
• Varia Meghal
• Gadelrab Serag
• Hsieh Cheng-Teh
• Podaima Jason Edward
• Szeto Victor
• Boisjoly Richard
• Aleksic Milivoje
• Longo Tom
• Chong In-Suk

Classifications

• H03M7/6023

  Parallelization · CPC title

• G06N5/04

  Inference or reasoning models · CPC title

• G06F13/28

  using burst mode transfer, e.g. direct memory access {DMA}, cycle steal (G06F13/32 takes precedence) · CPC title

• H03M7/3077 Primary

  Sorting · CPC title

• G06N20/00

  Machine learning · CPC title