On-Device Machine Learning Models for Real-Time Call Remediation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining voice call quality rely on network-side data, which provides a limited user-centric view, and on-device computing resources are insufficient for effective real-time analysis and remediation of call quality issues.
Innovation Solution
Implementing on-device machine learning models trained remotely to analyze user equipment data, allowing for real-time identification of call quality issues and execution of remedial actions using locally available resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If network-side data is used for determining voice call quality, then the analysis can be performed with sufficient computing resources, but the user-centric view and timeliness of call quality analysis are limited
Solution Approach 1:
The system pre-loads machine learning models onto the user equipment before they are needed for analysis. This preliminary action allows the models to be readily available when call quality issues occur, enabling immediate local analysis without waiting for model deployment or cloud processing, thus resolving the timeliness issue while maintaining analysis accuracy
Solution Approach 2:
Instead of the traditional approach where the network analyzes call quality data centrally, this patent inverts the architecture by enabling the user equipment to perform autonomous analysis using locally stored models. This inversion allows real-time analysis at the edge of the network, simultaneously achieving both high accuracy through sophisticated models and immediate timeliness through local processing
2Productivity
If machine learning models are deployed on user equipment, then real-time analysis and remediation can be performed, but the computing resources on user equipment are insufficient for effective analysis
Solution Approach 1:
The system implements a selective approach where only specific machine learning models relevant to particular call quality issues are loaded onto the user equipment, rather than deploying complete analysis systems. This partial action allows the device to perform targeted real-time analysis with minimal computing resource consumption, achieving productivity improvement without overwhelming the device's power constraints
Solution Approach 2:
The patent dynamically adjusts the complexity and size of machine learning models based on the available computing resources and the specific analysis task at hand. By changing model parameters such as model size, complexity, and computational requirements, the system can execute sophisticated analyses when resources are abundant while falling back to lighter models when resources are constrained, thus resolving the contradiction between real-time analysis capability and computing resource availability
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for utilizing an on-device machine learning model are disclosed. In one aspect, a method includes the actions of receiving a model that is trained using machine learning and that is configured to determine a given cause of a given event associated with a computing device that is executing the application. The actions further include accessing device data. The actions further include accessing network data. The actions further include providing the device data, the network data, and data identifying an event associated with the computing device as an input to the model. The actions further include receiving, from the model, data indicating a cause of the event. The actions further include determining an action that remediates the cause of the event. The actions further include performing the action that remediates the cause of the event.