Split Scheduler for AI/ML Task Offloading in 5G Terminals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The integration of artificial intelligence (AI) and machine learning (ML) in 5G mobile communication systems faces challenges due to compute-intensive and energy-intensive operations, which often exceed the resource limits of terminal and network side nodes, leading to incomplete handling of AI/ML operation-type services.
Innovation Solution
A communication method and device that send auxiliary information from network nodes to terminals to determine whether to initiate AI/ML operations, optimizing resource allocation by splitting compute-intensive tasks to network nodes while keeping privacy-sensitive and delay-sensitive tasks on terminals, using a split scheduler to manage resource allocation and bandwidth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If AI/ML operations are executed on terminal devices, then service responsiveness and privacy protection are improved, but device energy consumption and computational resource requirements increase beyond available limits
Solution Approach 1:
The patent segments AI/ML operations into two categories: privacy-sensitive and delay-sensitive tasks executed on terminals, and compute-intensive tasks executed on network side nodes. This segmentation allows the system to distribute computational workload appropriately, maintaining service responsiveness for time-critical operations while offloading energy-intensive computations to network infrastructure with sufficient resources.
Solution Approach 2:
The patent introduces a split scheduler as an intermediary component that manages resource allocation between terminal and network side nodes. The split scheduler receives service requests, evaluates resource availability, and dynamically determines which node should execute each AI/ML task based on current system state, thereby optimizing energy consumption while maintaining service quality.
2Adaptability or versatility
If compute-intensive AI/ML tasks are executed on terminal devices, then service customization and data privacy are improved, but device computational resources are insufficient to handle the operations
Solution Approach 1:
The patent segments AI/ML tasks based on their computational requirements and characteristics. Compute-intensive tasks that exceed terminal capabilities are routed to network side nodes, while tasks requiring service customization and data privacy are executed on terminals. This segmentation enables the system to handle diverse AI/ML workloads without overwhelming terminal device resources.
Solution Approach 2:
The patent extends the execution environment from a single dimension (terminal-only) to multiple dimensions by introducing network side nodes as additional execution platforms. This dimensional expansion allows the system to leverage network resources for compute-intensive operations while preserving terminal autonomy for privacy-sensitive tasks, effectively resolving the resource constraint through spatial distribution.
3Ease of operation
If all AI/ML operations are handled by network side nodes, then terminal device resource constraints are relieved, but network node resources become insufficient and service latency increases
Solution Approach 1:
The patent segments the AI/ML processing workload between terminal devices and network side nodes based on task characteristics. By identifying and isolating compute-intensive tasks for network execution while keeping other tasks on terminals, the system prevents network node resource exhaustion and reduces unnecessary network traffic, thereby maintaining high productivity and low latency.
Solution Approach 2:
The patent applies partial action by having terminals execute only the necessary portion of AI/ML tasks locally (privacy-sensitive and delay-sensitive operations), rather than offloading all operations to the network. This partial local execution reduces network node processing burden and communication overhead, preventing resource insufficiency and latency issues.
4Speed
If AI/ML services are initiated without resource evaluation, then service initiation speed is improved, but resource allocation efficiency deteriorates and service completion fails
Solution Approach 1:
The patent implements preliminary action through the split scheduler, which evaluates resource availability and determines task execution locations before AI/ML services are initiated. This pre-evaluation ensures that services are started only when appropriate resources are available, preventing failed completions while maintaining efficient resource allocation. The split scheduler maintains current resource state information to make rapid allocation decisions.
Data Source
AI summary
A communication method includes: sending first information, where the first information is used for enabling a terminal to determine whether to initiate an artificial intelligence or machine learning operation-type service.