AI Retrieval Pipeline Using Precomputed Features for Faster Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The processing speed of data generation using generation AI is hindered by the need to generate intermediate representation data each time, and there is a need to reduce the data capacity stored in storage areas.
Innovation Solution
An information processing method that includes generating features from input data, retrieving corresponding data from a pre-computed correspondence relationship table, and using this data to input to a generation AI, thereby reducing the need for real-time generation of intermediate representation data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If intermediate representation data is generated each time using a neural network, then data accuracy is maintained, but processing speed deteriorates
Solution Approach 1:
The patent pre-generates intermediate representation data for multiple input images before generation AI processing, storing them in a storage unit. This preliminary action eliminates the need for real-time neural network processing during generation AI execution, significantly improving processing speed while maintaining data accuracy through pre-computed high-quality intermediate representations.
Solution Approach 2:
The patent extracts the intermediate representation generation step from the real-time processing pipeline and performs it separately in advance. By separating the intermediate representation generation from the generation AI processing, the system maintains accurate intermediate data while enabling faster overall processing throughput.
2Quantity of substance
If data capacity stored in storage area is reduced through compression, then storage efficiency improves, but processing complexity increases
Solution Approach 1:
The patent applies different processing strategies to different data based on their characteristics. Important intermediate representation data that benefits from compression is stored in the storage unit with compressed formats, while maintaining the ability to retrieve and process it efficiently. This localized application of compression to specific data types reduces overall storage requirements without unnecessarily complicating the entire processing system.
Data Source
AI summary
In an information processing method, an information processing system generates a feature of input data, and acquires retrieved data of the input data corresponding to the feature, based on correspondence relationship information between the feature and the retrieved data. The information processing system inputs the acquired retrieved data to a generation artificial intelligence (AI), and acquires answer data to the input data from the generation AI.