Data compression by using cognitive created dictionaries

• US 10,838,922 B2
• Filed: 03/31/2017
• Issued: 11/17/2020
• Est. Priority Date: 03/31/2017
• Status: Active Grant

• Current Assignee: Beijing Zitiao Network Technology Company Limited (ByteDance Ltd.)
• Original Assignee: International Business Machines Corporation
• Inventors: Pankanti, Sharathchandra Umapathirao, Rueger, Erik, Schaefer, Rolf
• Primary Examiner(s): Rostami, Mohammad S
• Application Number:
  US15/475,939
• Publication Number:
  US 20180285378A1
• Time in Patent Office: 1,327 Days
• CPC Class Codes: G06F 16/1744 using compression, e.g. sparse files
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 5/045 Explanation of inference; Explainable artificial intelligence [XAI]; Interpretable artificial intelligence

Abstract

A compression method, system, and computer program product include creating compressed data via a first system from input data, sending information to a second system detailing a compression strategy for the compressed data, and learning, via the second system, from the information how to recreate the input to the first system using the compressed data.

Claims

1. A computer-implemented compression method, the method comprising:

creating compressed data via a first system from input data;

sending information to a second system detailing a compression strategy for the compressed data;

learning, via the second system, from the information how to recreate the input to the first system using the compressed data;

decompressing, via the second system, the compressed data to a recreated input;

comparing the input data to the first system with the recreated input in an iterative loop;

based on a result of the comparing and the iterative loop of the comparing, modifying the information at each iterative loop how to recreate the input such that the input data to the first system matches the recreated input; and

sending a feedback to the first system based on the recreated input by the second system,

wherein the first system teaches the second system how to recreate the input simultaneously while the second system teaches the first system an effectiveness of the teaching that the first system provides to the second system,

further comprising measuring a quality of the recreated input by the second system to send a feedback to the first system to adjust the information for the compression strategy,

wherein the information is not sent to the second system with a next compressed data when the quality is greater than a predetermined threshold value,

wherein the first system and the second system are trained by the learning and feedback without training data and/or pre-annotated data,

wherein the input data comprises a set of unstructured data of special types having multiple categories to classify each of the special types into one of the multiple categories, and

wherein the first system and the second system comprise a neuronal network or a cognitive expert.

2. The computer-implemented method of claim 1, further comprising:

measuring a size of the compressed data; and

if the compression strategy is modified, repeating the creating, the learning, and the sending to confirm that the modified compression strategy creates compressed data including a minimum size by comparing a new size to the measured size.

3. The computer-implemented method of claim 1, wherein the compressed data is created with a minimum size with an effort level of decoding, the effort level of decoding is dependent on a number of computing steps of the second system.

4. The computer-implemented method of claim 1, wherein the feedback is sent to the first system if the recreated input and the input do not match.

5. The computer-implemented method of claim 1, further comprising:

if the compression strategy is modified, repeating the creating, the learning, and the sending to confirm that the modified compression strategy creates compressed data including a minimum size by comparing a new size to the measured size,

wherein the information is not sent to the second system with a next compressed data when the quality is greater than a predetermined threshold value.

6. The computer-implemented method of claim 1, further comprising:

measuring a size of the compressed data; and

if the compression strategy is modified, repeating the creating, the learning, and the sending to confirm that the modified compression strategy creates compressed data including a minimum size by comparing a new size to the measured size,

wherein the compressed data is created with a minimum size with an effort level of decoding, the effort level of decoding is dependent on a number of computing steps of the second system, and

wherein the feedback is sent to the first system if the recreated input and the input do not match.

7. A computer program product for compression, the computer program product comprising a computer-readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform:

creating compressed data via a first system from input data;

sending information to a second system detailing a compression strategy for the compressed data;

learning, via the second system, from the information how to recreate the input to the first system using the compressed data;

decompressing, via the second system, the compressed data to a recreated input;

comparing the input data to the first system with the recreated input in an iterative loop;

based on a result of the comparing and the iterative loop of the comparing, modifying the information at each iterative loop how to recreate the input such that the input data to the first system matches the recreated input; and

sending a feedback to the first system based on the recreated input by the second system,

wherein the first system teaches the second system how to recreate the input simultaneously while the second system teaches the first system an effectiveness of the teaching that the first system provides to the second system,

further comprising measuring a quality of the recreated input by the second system to send a feedback to the first system to adjust the information for the compression strategy,

wherein the information is not sent to the second system with a next compressed data when the quality is greater than a predetermined threshold value,

wherein the first system and the second system are trained by the learning and feedback without training data and/or pre-annotated data,

wherein the input data comprises a set of unstructured data of special types having multiple categories to classify each of the special types into one of the multiple categories, and

wherein the first system and the second system comprise a neuronal network or a cognitive expert.

8. The computer program product of claim 7, further comprising:

measuring a size of the compressed data; and

if the compression strategy is modified, repeating the creating, the learning, and the sending to confirm that the modified compression strategy creates compressed data including a minimum size by comparing a new size to the measured size.

9. The computer program product of claim 7, wherein the compressed data is created with a minimum size with an effort level of decoding, the effort level of decoding is dependent on a number of computing steps of the second system.

10. The computer program product of claim 7, wherein the recreated input by the second system is compared with the input to the first system, and

the feedback is sent to the first system if the recreated input and the input do not match.

11. A compression system, said system comprising:

a processor; and

a memory, the memory storing instructions to cause the processor to perform;

creating compressed data via a first system from input data;

sending information to a second system detailing a compression strategy for the compressed data;

learning, via the second system, from the information how to recreate the input to the first system using the compressed data;

decompressing, via the second system, the compressed data to a recreated input;

comparing the input data to the first system with the recreated input in an iterative loop;

based on a result of the comparing and the iterative loop of the comparing, modifying the information at each iterative loop how to recreate the input such that the input data to the first system matches the recreated input; and

sending a feedback to the first system based on the recreated input by the second system,

wherein the first system teaches the second system how to recreate the input simultaneously while the second system teaches the first system an effectiveness of the teaching that the first system provides to the second system,

further comprising measuring a quality of the recreated input by the second system to send a feedback to the first system to adjust the information for the compression strategy,

wherein the information is not sent to the second system with a next compressed data when the quality is greater than a predetermined threshold value,

wherein the first system and the second system are trained by the learning and feedback without training data and/or pre-annotated data,

wherein the input data comprises a set of unstructured data of special types having multiple categories to classify each of the special types into one of the multiple categories, and

wherein the first system and the second system comprise a neuronal network or a cognitive expert.

12. The system of claim 11, embodied in a cloud-computing environment.