Edge ML Model Retraining via Automated Signal Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional edge computing systems face challenges in handling large amounts of data from industrial machines due to limitations in connectivity, bandwidth, and the need for real-time decision-making and predictive maintenance, especially in environments where human intervention for data labeling is impractical.
Innovation Solution
A method for supervised learning that enables automatic retraining of machine learning models at edge devices without human intervention, using positive and negative signals generated in a deployed environment, supporting various sensor types and allowing for automatic detection of events and model retraining based on data distribution changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If data is sent to cloud storage for processing, then centralized computing power is available, but connectivity requirements increase and real-time decision-making is delayed
Solution Approach 1:
The system segments the computing architecture into edge devices that perform local real-time processing and cloud data centers that handle non-time-critical tasks. The edge device processes sensor data locally to enable real-time predictive maintenance decisions, while cloud storage handles historical data archiving and model retraining, thus resolving the contradiction between centralized computing power and real-time response requirements
Solution Approach 2:
The edge device acts as an intermediary between the cloud data center and the industrial machines. It receives data from machines, performs local analysis for immediate decisions, and only transmits processed results or selected raw data to the cloud, thereby reducing latency for time-critical operations while maintaining the benefits of centralized computing resources
2Loss of information
If all sensor data is transmitted to cloud storage, then data availability is improved, but bandwidth requirements increase and costs become prohibitive
Solution Approach 1:
The system extracts only the essential information from raw sensor data at the edge device level. The edge device processes and analyzes data locally, extracting key features and insights, then transmits only these extracted insights rather than the complete raw data sets to the cloud, thereby maintaining data availability while dramatically reducing bandwidth consumption and associated costs
Solution Approach 2:
The data transmission is segmented into different categories: time-critical data processed and decided locally at the edge, non-time-critical aggregated data transmitted periodically to the cloud, and only essential alerts or anomalies sent immediately. This segmentation reduces overall bandwidth requirements while ensuring critical information remains available
3Measurement precision
If human intervention is used for data labeling in supervised learning, then model accuracy is improved, but the process becomes impractical in deployed environments
Solution Approach 1:
The system implements self-service through automated model retraining at the edge device. The edge device automatically collects new data during operation, performs local analysis to identify labeling opportunities, and retrains models using this data without requiring human intervention. This maintains model accuracy through continuous adaptation while making the system operationally practical in deployed environments
Solution Approach 2:
The system performs preliminary data collection and analysis at the edge device before needing human labeling. By pre-processing data, identifying relevant patterns, and preparing training datasets in advance, the system reduces the burden of human labeling while maintaining model accuracy through automated retraining cycles
4Adaptability or versatility
If model retraining is performed continuously, then model accuracy adapts to changing conditions, but computational resources and time are consumed
Solution Approach 1:
The system implements periodic model retraining at the edge device based on accumulated data thresholds or time intervals rather than continuous retraining. The edge device monitors data accumulation and triggers retraining only when sufficient new data is available or at scheduled intervals, enabling model adaptation to changing conditions while conserving computational resources and energy
Solution Approach 2:
The system performs partial retraining by focusing computational resources on retraining only with the most recent and relevant data subsets rather than the entire historical dataset. This selective retraining approach maintains model adaptability to current conditions while significantly reducing the computational burden compared to full dataset retraining
Data Source
AI summary
A set of processes enable supervised learning of a machine learning model without human intervention by producing the positive and negative examples at-will in a deployed environment. A technique implements a series of events that replaces the need for human intervention to generate labeled data for supervised learning. This enables automatic retraining of the model in a deployed environment without the need for human labeled data, supporting audio and video data.