Computational Storage for Neural Multimedia Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The performance of electronic devices is limited by inefficient data transmission speeds between host devices and storage devices, particularly when managing high-capacity multimedia data, which can lead to a lack of storage space and deteriorated system performance.
Innovation Solution
A computational storage system that includes a storage device and a computing device, utilizing neural network models to generate inference multimedia data based on base data and event tables, allowing efficient storage of reduced-capacity data representations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If multimedia data is stored in the storage device, then data storage capacity is utilized, but storage space efficiency deteriorates due to high-capacity data requirements
Solution Approach 1:
The original multimedia data is segmented into multiple components: base data (essential information), event data (changes occurring in the data), and inference data (regenerated from base and event data). This segmentation allows storing only the compact base data and event data instead of the complete original data, significantly improving storage space efficiency while preserving data integrity and functionality.
2Productivity
If data transmission speed between host device and storage device is increased, then throughput improves, but system performance is still limited by the transmission bottleneck
Solution Approach 1:
The invention extracts and stores only the essential base data and event data from the complete multimedia data. By removing redundant information and storing only the minimal necessary data components, the system reduces the amount of data that needs to be transmitted between storage and host devices, thereby improving throughput without being constrained by transmission speed limitations.
3Quantity of substance
If neural network models are used to generate inference multimedia data, then storage space efficiency is improved, but device complexity increases
Solution Approach 1:
The neural network model is trained in advance during an offline phase to learn the mapping relationships between base data, event data, and the original multimedia data. This preliminary action allows the model to be saved and reused for generating inference data without requiring complex real-time processing during data operations, thus reducing operational complexity while maintaining storage efficiency benefits.
Data Source
Figure 1~2
Figure 3A
Figure 3B
AI summary
An example computational storage system includes a storage device (210) and a computing device (220). The computing device (220) is configured to generate first inference multimedia data corresponding to original multimedia data based on base data, an event table, and at least one neural network model, where the base data includes base raw data of at least one object included in the original multimedia data, and the event table includes data obtained based on respectively mapping events occurred in the base data and occurrence times of the events.