Hierarchical AI Computing System for Inference Precision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI computing systems face challenges in integrating multiple frameworks quickly due to differing hardware specifications, leading to lengthy development cycles, complex data transmission, and difficulties in function integration and maintenance.
Innovation Solution
A hierarchical AI computing system with multiple layers of AI subsystems, where each layer performs inference based on internal and external data, using standardized AI model description files to facilitate rapid implementation and reduce data transmission, allowing for adaptive planning and efficient data processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If centralized AI algorithms are used to transmit all operational data to a specified computing device for analysis, then comprehensive data analysis can be achieved, but data transmission costs increase and computation time increases
Solution Approach 1:
The patent divides the centralized AI computing system into multiple hierarchical layers with distributed AI subsystems. Each layer processes data locally before transmitting only essential results upward, segmenting the data transmission workload and reducing overall transmission costs while maintaining comprehensive analysis capability.
Solution Approach 2:
The patent introduces a hierarchical dimension to the previously flat centralized system. By organizing AI subsystems across multiple layers (edge, fog, cloud), the system adds a vertical dimension that enables local processing at lower layers and aggregate analysis at higher layers, reducing the need to transmit all raw data to a single central point.
2Reliability
If AI models are customized for each electronic device based on hardware specifications, then device-specific optimization is achieved, but development cycles lengthen and maintenance becomes difficult
Solution Approach 1:
The patent creates a universal AI model description file format and standardized interface that can be used across different electronic devices with varying hardware specifications. This universal framework allows the same AI model to be deployed on multiple device types without customization, while still achieving device-specific optimization through adaptive inference at the edge layer.
Solution Approach 2:
The patent enables AI models to adapt to different hardware specifications by changing inference parameters rather than retraining the models. The standardized AI model description files contain configurable parameters that can be adjusted based on device capabilities, allowing the same model architecture to operate efficiently across diverse hardware without lengthy retraining processes.
3Adaptability or versatility
If multiple AI frameworks are integrated into a single system, then framework compatibility is improved, but system complexity increases and integration difficulties arise
Solution Approach 1:
The patent introduces a standardized AI model description file format as an intermediary layer between different AI frameworks and the hierarchical AI subsystems. This intermediary standard enables multiple frameworks to communicate through a common interface, reducing integration complexity while maintaining framework compatibility and versatility.
4Measurement precision
If data sampling rate is increased to improve feature extraction efficiency, then inference precision improves, but data transmission costs increase
Solution Approach 1:
The patent extracts only the essential inference results from each hierarchical layer rather than transmitting all raw data. By taking out only the processed information needed for higher-layer decision-making, the system maintains high inference precision through increased local sampling while dramatically reducing transmission costs by sending only condensed results.
Data Source
AI summary
A hierarchical artificial intelligence (AI) computing system includes at least one group of a first layer AI subsystems and n second layer AI subsystems. One of the at least one group of a first layer AI subsystems includes m first layer AI subsystems, and each of the m first layer AI subsystems is configured to perform inference based on internal sensing data or a first external sensing data to generate a first inference result; and the n second layer AI subsystems are respectively connected to the at least one group of the first layer AI subsystems, where each of the n second layer AI subsystems is configured to perform inference based on m first inference results, an operation command, and a second external sensing data to generate a second inference result.