Confidence-Based Model Verification for IoT Edge Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In IoT applications, the varying perception capabilities and performance differences between edge devices and core devices make it challenging to implement a stacked machine learning strategy across different devices effectively.
Innovation Solution
A method where a first device acquires features associated with a target object, determines a confidence coefficient using a trained model, and sends requests to a second device with a more complex model for verification, allowing the first model to be updated based on the response, thereby implementing a stacked machine learning strategy that coordinates the use of perception capabilities and computing resources across devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a more complex model is deployed at the second device for verification, then model accuracy is improved, but device complexity and computing costs increase
Solution Approach 1:
The verification system is segmented into multiple devices (first device and second device), each deploying models of appropriate complexity for their specific roles. The first device uses a simpler model for initial assessment, while the second device uses a more complex model for verification only when needed, distributing the computational burden and accuracy requirements across the system architecture.
Solution Approach 2:
Different devices are assigned different model complexities based on their local capabilities and roles in the verification workflow. The second device, which handles verification tasks, is equipped with a more complex model to ensure high accuracy, while the first device uses a simpler model for routine operations, optimizing the balance between accuracy and complexity at each local level.
2Measurement precision
If a more complex model is used for verification, then model accuracy is improved, but computing costs increase
Solution Approach 1:
The complex model at the second device is not continuously executed but only activated partially when verification is required based on confidence coefficient thresholds. This partial action approach ensures high accuracy when needed while avoiding the excessive computing costs of continuously running the complex model, optimizing the trade-off between accuracy and energy consumption.
Solution Approach 2:
The system implements self-service by automatically determining when verification is needed based on confidence coefficients from the first model. The second device with the complex model only processes cases that require verification, allowing the system to serve itself efficiently without manually triggering expensive computations for every input.
3Productivity
If confidence threshold verification is implemented, then decision-making efficiency is improved, but additional verification requests increase communication overhead
Solution Approach 1:
The system uses feedback from the first model's confidence coefficient to dynamically determine whether verification is needed. When the confidence coefficient exceeds the threshold, the system efficiently makes decisions without additional communication. When it falls below the threshold, feedback triggers a verification request to the second device, optimizing the balance between decision-making efficiency and communication overhead through intelligent feedback-driven routing.
Data Source
AI summary
Embodiments include a method, an electronic device, and a computer program product for information processing. In an example embodiment, a method for information processing includes: acquiring, at a first device, a first feature associated with a target object; applying the first feature to a trained first model deployed at the first device to determine a first confidence coefficient, the first confidence coefficient being associated with probabilities that the first model determines the target object as a real object and as a false object; if the first confidence coefficient is lower than a first threshold confidence coefficient, sending a request for verifying the target object to a second device, the second device being deployed with a trained second model for verifying the target object, and the second model being more complex than the first model; and updating the first model based on a response to the request.