Edge Device ML Model Management via Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated machine learning (autoML) methods prioritize model accuracy over execution time on resource-constrained devices, leading to suboptimal performance in scenarios like real-time video object detection on edge devices, where both accuracy and computational efficiency are critical.
Innovation Solution
The method involves generating and training a set of specialized machine learning models for different situations using an autoML engine, with a monitoring and decision module selecting the most appropriate model for the resource-constrained device based on input data characteristics and execution context, allowing for optimized model accuracy and reduced computation time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single high-accuracy machine learning model is deployed on a resource-constrained edge device, then model accuracy is improved, but execution time and computational resource consumption increase
Solution Approach 1:
The patent segments the machine learning model into multiple specialized models, each optimized for specific input data characteristics or situations. The monitoring and decision module divides the input space and routes different inputs to appropriate specialized models, allowing the system to achieve high accuracy for each segment while maintaining fast execution by selecting the most appropriate model for each input rather than using a single large model for all cases.
2Loss of time
If multiple specialized models are maintained and switched between, then execution time for specific inputs is reduced, but device complexity increases
Solution Approach 1:
The patent introduces a monitoring and decision module as an intermediary between the input data and the multiple specialized models. This module monitors input data characteristics, determines the appropriate situation or input space region, and selects which specialized model to activate. This intermediary manages the complexity by providing a systematic approach to model selection and switching, preventing the need for complex ad-hoc model management while enabling fast execution through appropriate model selection.
3Speed
If model switching logic is implemented on the edge device, then execution speed is improved, but memory and processing requirements increase
Solution Approach 1:
The patent performs preliminary actions by pre-defining input space regions and associating each region with appropriate specialized models during the model training and deployment phase. The monitoring and decision module uses these pre-established associations to quickly route inputs to the correct model without requiring complex real-time analysis or switching logic. This preliminary organization of models and input spaces reduces the computational and memory burden during actual execution while maintaining fast response times.
Data Source
AI summary
A method of solving a machine learning (ML) problem using a resource-constrained device includes generating and training, by an automated machine learning (autoML) engine, a model set including a number of different models for the ML problem. Each of the different models of the model set is specialized for a particular situation. The method further includes monitoring, by a monitoring and decision module, input data of the ML problem and selecting one or more models of the model set as active models to be applied by the resource-constrained device. The method also includes receiving, by the resource-constrained device, input data of the ML problem and applying the one or more models selected by the monitoring and decision module to the received input data.