Task Processing Model Confidence Escalation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex machine learning models are difficult to deploy in devices with limited computing resources, such as mobile devices, due to high demands for computing resources, leading to a balance issue between model processing accuracy and efficiency.
Innovation Solution
A method where a target task is processed by a first device using a deployed first model, and if the confidence of the result is lower than a threshold, the task is processed by a second device using a more complex model deployed on a device with greater computing power, where the first model is constructed by compressing the second model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a complex machine learning model is deployed to improve processing accuracy, then model processing accuracy is improved, but computing resource consumption increases and deployment difficulty increases for devices with limited resources
Solution Approach 1:
The patent divides the model processing system into multiple devices with different computing capabilities. Complex models are deployed on powerful devices (e.g., servers) while simpler versions or model outputs are used on devices with limited resources (e.g., mobile devices). This segmentation allows each device to operate within its resource constraints while collectively achieving high processing accuracy.
Solution Approach 2:
The patent introduces an intermediary mechanism where a first device processes tasks using a first model and obtains results with confidence scores. When confidence is insufficient, the system intermediates by transferring the task to a second device with greater computing power that has a more complex model deployed. This intermediary approach balances resource usage across devices while maintaining accuracy.
2Measurement precision
If a complex machine learning model is deployed to improve processing accuracy, then model processing accuracy is improved, but processing efficiency decreases due to higher computing resource demands
Solution Approach 1:
The patent implements partial action by having the first device attempt to process tasks locally first using a lighter model. Only when the confidence score falls below a threshold does the system invoke the more resource-intensive second device. This partial approach to using complex models minimizes overall computing resource consumption while maintaining accuracy for difficult cases.
Solution Approach 2:
The confidence score mechanism acts as an intermediary that determines when to escalate from local processing to remote processing. This intermediary layer allows the system to efficiently decide whether the benefits of using a complex model justify the additional computing resource consumption, thereby optimizing the balance between accuracy and efficiency.
3Measurement precision
If model complexity is increased to improve processing accuracy, then model processing accuracy is improved, but deployment difficulty increases for devices with limited computing resources
Solution Approach 1:
The patent segments the deployment architecture into multiple devices with different capabilities. Highly complex models are deployed only on devices with sufficient resources (e.g., servers), while simpler models or model inference capabilities are deployed on devices with limited resources (e.g., mobile devices). This segmentation makes deployment feasible across diverse device types while maintaining the ability to achieve high accuracy when needed.
Solution Approach 2:
The system creates a universal processing framework that can handle both simple and complex model processing across different devices. The first device can process tasks locally using a lightweight model, while the second device provides backup complex model processing capability. This multi-functional architecture allows the same system to serve multiple device types with varying resource constraints.
Data Source
AI summary
Embodiments of the present disclosure relate to a method, an electronic device, and a computer program product for task processing. The method includes: processing, in response to receiving a target task, the target task by a first device using a deployed first model; acquiring a first result determined by the first model, the first result having a first confidence; processing, in response to determining that the first confidence is lower than a first threshold, the target task by a second device using a deployed second model; and acquiring a second result determined by the second model, the first model being constructed by compressing the second model. In this way, the accuracy of task processing can be ensured.