Multi-task Deployment for Deep Learning Task Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning systems face challenges in matching suitable models to tasks in real-time, especially when complex tasks change, leading to difficulties in ensuring schedulability and efficiency.
Innovation Solution
A multi-task deployment method that allocates tasks to network models to generate candidate combinations, selects the combination with maximum operation accuracy, and deploys the target mapping relation on a prediction machine, using algorithms like PSO and EDF to optimize task processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If deep learning models are configured based on experience, then the system can operate with existing models, but it cannot ensure real-time schedulability when complex task changes occur
Solution Approach 1:
The patent implements dynamic task allocation by continuously monitoring task characteristics and matching them with appropriate network models in real-time. The system dynamically adjusts the mapping between tasks and models based on current task requirements, rather than using static experience-based configuration. This dynamic approach ensures both adaptability to task changes and reliability through optimized resource allocation.
Solution Approach 2:
The system employs feedback mechanisms by evaluating task execution results and performance metrics to continuously improve task-model matching. The feedback loop allows the system to learn from past performance and adjust future allocations, ensuring both adaptability to new tasks and reliability in meeting scheduling requirements through data-driven optimizations.
2Adaptability or versatility
If multiple network models are deployed to handle various tasks, then task coverage is improved, but system complexity increases
Solution Approach 1:
The patent implements a universal task allocation framework that can manage multiple network models through a single standardized interface and matching mechanism. This framework provides multi-functional capabilities by handling diverse task types, model selections, and deployment configurations through unified algorithms, thereby improving task coverage while controlling system complexity through standardization.
Solution Approach 2:
The system manages complexity by dynamically adjusting parameters such as task priorities, model selection criteria, and allocation strategies based on current system state and task characteristics. By changing these parameters rather than restructuring the entire system, the patent achieves versatile task coverage while maintaining manageable system complexity through flexible parameter optimization.
3Measurement precision
If task allocation is optimized for accuracy, then operation accuracy improves, but processing time increases
Solution Approach 1:
The patent applies partial optimization by allocating computational resources selectively based on task requirements. Rather than optimizing all tasks to maximum accuracy, the system applies enhanced accuracy optimization only where necessary, while using faster, less resource-intensive models for tasks where high accuracy is less critical. This partial action approach improves overall accuracy while minimizing additional processing time.
Solution Approach 2:
The system dynamically changes accuracy parameters based on task priorities and deadlines. For time-sensitive tasks, the system adjusts accuracy requirements to acceptable levels that meet timing constraints, while for non-time-critical tasks, it optimizes for maximum accuracy. This parameter adjustment strategy balances accuracy improvements with time efficiency across different task types.
Data Source
AI summary
The disclosure provides a multi-task deployment method, and an electronic device. The method includes: obtaining N first tasks and K network models, in which N and K are positive integers greater than or equal to 1; allocating the N first tasks to the K network models differently for operation, to obtain at least one candidate combination of tasks and network models, in which each candidate combination includes a mapping relation between the N first tasks and the K network models; selecting a target combination with a maximum combination operation accuracy from the at least one candidate combination; and deploying a target mapping relation comprised in the target combination and the K network models on a prediction machine.