Distributed AI Model Processing via Submodel Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
On-device AI processing is limited by the hardware capabilities of electronic devices, restricting the performance of artificial intelligence models and compromising user privacy and data processing speed.
Innovation Solution
A method for distributed processing of artificial intelligence models, where the model is divided into submodels that can be processed in parallel, with appropriate processing devices determined based on computation requirements and predicted processing times, allowing for efficient resource allocation across multiple devices within a local network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If on-device AI processing is used to protect user privacy and maintain data processing speed, then data security and processing speed are improved, but processing performance is limited by hardware capabilities
Solution Approach 1:
The AI model is divided into multiple submodels that can be distributed across different processing devices. Each submodel is processed by a separate device in parallel, allowing the system to overcome the hardware limitations of a single device while maintaining data security within the local network.
Solution Approach 2:
Multiple processing devices are combined to form a distributed processing system. The capabilities of individual devices are merged through parallel processing of submodels, achieving superior processing performance while keeping all operations within the secure local network environment.
2Device complexity
If the AI model is processed by a single device to maintain simplicity, then device complexity is reduced, but processing speed and performance are limited
Solution Approach 1:
The AI model is segmented into multiple submodels that can be independently processed. This segmentation enables parallel processing across multiple devices, significantly improving processing speed while maintaining manageable complexity through modular architecture.
Solution Approach 2:
The system dynamically assigns submodels to different processing devices based on their capabilities and current workload. This dynamic allocation optimizes processing speed while keeping the system architecture flexible and adaptable to changing conditions.
3Productivity
If the AI model is divided and processed in parallel across multiple devices, then processing performance is improved, but coordination and resource allocation become more complex
Solution Approach 1:
The system performs preliminary actions by pre-dividing the AI model into submodels and pre-assigning them to appropriate processing devices based on their capabilities. This preliminary organization simplifies coordination during actual processing, as the framework for parallel execution is established in advance.
Solution Approach 2:
The system implements feedback mechanisms to monitor the processing status of submodels across different devices. Based on this feedback, the system can dynamically adjust resource allocation and coordination strategies, optimizing performance while managing complexity through real-time information about system state.
Data Source
AI summary
A method for distributed processing of an artificial intelligence model includes: obtaining an artificial intelligence model to be distributed and processed, dividing the artificial intelligence model to obtain a plurality of submodels processable in parallel, determining a first processing device to process a first submodel requiring a greatest amount of computations from among the plurality of submodels, predicting a first processing time required for the first processing device to process the first submodel, determining a second processing device to process a second submodel from among the plurality of submodels based on the predicted first processing time, and performing the distributed processing of the artificial intelligence model by requesting the first processing device and the second processing device to process the first submodel and the second submodel, respectively.