Multi-Neural Network Scheduling for Uniform Output Frame Ratios
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automatic driving systems face challenges in processing multiple neural networks simultaneously due to finite acceleration processing units, leading to non-uniform output frame rates that fail to satisfy preset ratios, resulting in inconsistent processing results between neural networks.
Innovation Solution
Divide operator operation processes for each neural network into multiple executions according to a preset ratio of output frame rates and execute these processes sequentially among the networks to ensure uniform output and satisfy the desired frame rate ratios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple neural networks are processed simultaneously with finite acceleration processing units, then processing throughput is improved, but output frame rate uniformity deteriorates
Solution Approach 1:
The patent segments the processing of multiple neural networks into divided execution units. Each neural network's operator operation process is divided into multiple segments that can be executed in an interleaved manner across available acceleration processing units. This segmentation allows simultaneous processing while maintaining controlled output timing through the division and scheduling mechanism.
2Productivity
If finite acceleration processing units are used to process multiple neural networks, then resource utilization is improved, but output frame rate ratio satisfaction deteriorates
Solution Approach 1:
The patent implements dynamic scheduling of neural network processing tasks. The system dynamically adjusts the execution order and timing of operator operation processes based on the preset output frame rate ratios. This dynamic control mechanism ensures that despite using finite acceleration processing units, the system can still satisfy the required output frame rate ratios by flexibly managing resource allocation and task scheduling.
3Stability of the object's composition
If operator operation processes are executed sequentially by switching among networks, then output uniformity is improved, but processing time increases
Solution Approach 1:
The patent maintains continuity of useful action by implementing an interleaved execution strategy. Instead of completing one entire neural network processing task before starting another, the system switches between multiple neural networks at the operator operation process level, keeping acceleration processing units continuously occupied. This approach reduces idle time while maintaining output uniformity through controlled switching and division of execution units.
Data Source
AI summary
The present disclosure provides an output method for multiple neural networks. The method includes dividing an operator operation process for each of the neural networks or operator operation processes for part of the neural networks into multiple times of executions according to a preset ratio of output frame rates among the multiple neural networks; and executing the operator operation processes for the multiple neural networks sequentially by switching among the networks, such that the multiple neural networks output uniformly and satisfy the preset ratio of output frame rates.