Certifying Node for Edge ML Model Quality Assurance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning techniques face privacy and security concerns due to the need to share user data with central servers, leading to slow learning and large file sizes, which can result in data leaks and unauthorized interference.
Innovation Solution
A certifying node in a peer-to-peer network is used to incrementally train machine learning models, allowing edge devices to generate and certify predictive models locally without sharing sensitive data, using encrypted keys and Root Mean Square Error (RMSE) for quality assessment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If user data is sent to a central server for machine learning training, then the ML model can be trained using aggregated data, but privacy and security concerns arise due to potential data leaks and unauthorized access
Solution Approach 1:
The patent extracts the training data from the central server environment and processes it locally at edge devices. Only model parameters and gradients are transmitted to the server, while sensitive raw data remains distributed at the edge, eliminating the privacy and security risks associated with centralizing user data.
Solution Approach 2:
The system segments the centralized training architecture into distributed edge training nodes. Each edge device independently trains local model copies using its own data, and only shares aggregated gradients or parameter updates with the server, thereby segmenting the data flow to prevent centralized data accumulation and associated security risks.
2Productivity
If data is aggregated periodically at a central server, then the ML model can be updated, but the learning process becomes slow and not continuous
Solution Approach 1:
The patent implements continuous incremental training at edge devices where models are updated in real-time as new data arrives. This continuous local training eliminates the periodic batch processing delays inherent in centralized aggregation, allowing the system to maintain up-to-date models without waiting for scheduled data collection cycles.
Solution Approach 2:
Edge devices perform preliminary training actions locally before any server communication is needed. By pre-training models continuously at the edge and only synchronizing periodic updates or requesting corrections from the server when necessary, the system eliminates waiting time and maintains continuous learning productivity.
3Adaptability or versatility
If a trained ML model is transmitted repeatedly to the server, then the model can be updated and redistributed, but large file sizes impose significant network load
Solution Approach 1:
The patent extracts only the essential model parameters and gradient information needed for updates, transmitting only these compressed representations rather than complete model files. This extraction approach maintains model adaptability while dramatically reducing the quantity of data transmitted over the network.
Solution Approach 2:
The model update process is segmented into distributed incremental updates at the edge rather than monolithic model file transmissions. Each edge device independently updates its local model copy using small gradient updates or parameter adjustments, avoiding the need to repeatedly transmit large complete model files across the network.
4Productivity
If a candidate ML model is incrementally trained at an edge device, then real-time learning is achieved, but the model quality may be insufficient without proper certification
Solution Approach 1:
The patent implements a feedback mechanism where the certifying node evaluates candidate models using test datasets and provides quality assessments back to edge devices. This feedback loop ensures that incrementally trained models meet quality thresholds before deployment, maintaining reliability while preserving real-time learning productivity.
Solution Approach 2:
A certifying node is introduced as an intermediary between edge training and production deployment. This intermediary evaluates candidate models using test data, compares performance against thresholds, and certifies or rejects models accordingly, ensuring quality assurance without directly interfering with the real-time incremental training process at edge devices.
Data Source
AI summary
There is provided a method of operating a certifying node to certify incremental trained machine learning (ML) models of one or more edge devices associated with a peer to peer network. The method includes (i) generating a predictive outcome value for a test data set by executing a candidate ML model against the test data set available to the certifying node; (ii) determine a measure of quality of the candidate ML model by matching the predictive outcome value of the candidate ML model with an actual outcome value of the test data set; and (iii) certify the candidate ML model by comparing the measure of quality of the candidate ML model against a threshold error value, for use in real time incremental training by the one or more edge devices of the peer to peer network.