AI Model Compression by Data Distribution Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current memory systems struggle to efficiently store and process large artificial intelligence (AI) models due to high demands for data movement bandwidth and storage capacity, leading to latency and throughput issues.
Innovation Solution
The proposed solution involves categorizing AI model data based on its distribution analysis, selecting an appropriate compression algorithm for each category, and storing the compressed data in a solid-state drive. This approach includes generating address boundary information and storing a mapping between this information and the compression algorithm to facilitate efficient decompression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data is compressed using a single compression algorithm, then the storage capacity is reduced, but the compression ratio is limited and cannot be optimized for different data types
Solution Approach 1:
The patent segments AI model data into different categories based on distribution characteristics (e.g., uniform distribution, non-uniform distribution). Each category is then assigned a specific compression algorithm optimized for its characteristics, enabling higher overall compression ratios while efficiently reducing storage capacity requirements.
Solution Approach 2:
The patent applies different compression algorithms to different segments of data based on their local characteristics. By analyzing the distribution of data in specific regions or categories and applying tailored compression techniques, the system achieves optimized compression ratios for each data type while collectively reducing total storage needs.
2Productivity
If compression algorithms are selected based on data distribution analysis, then the compression ratio is improved, but the system complexity increases due to categorization and algorithm selection
Solution Approach 1:
The patent performs preliminary analysis of data distribution characteristics before compression. By categorizing data into different distribution types in advance and pre-selecting appropriate compression algorithms for each category, the system simplifies the overall process while maintaining high compression ratios, avoiding complex real-time decision-making during compression.
3Speed
If AI model data is stored without compression, then the access speed is maintained, but the storage bandwidth and data movement requirements become excessive
Solution Approach 1:
The patent changes the state of data by applying compression algorithms that transform data representation based on its distribution characteristics. This parameter change reduces the volume of data stored and moved, thereby reducing storage bandwidth requirements and data movement overhead, while decompression during access maintains effective access speed for AI workloads.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided are systems, methods, and apparatuses for compression of artificial intelligence models. In one or more examples, the systems, devices, and methods include categorizing data based on an analysis of a distribution of the data, generating compressed data based on the data and on a compression algorithm that is selected based on the categorization, and storing the compressed data in a storage device. In one or more examples, the systems, devices, and methods include identifying an address associated with compressed data based on a request for the compressed data, determining a decompression algorithm based on the address, and decompressing the compressed data using the determined decompression algorithm.