Multi-Model System for Wearable Sensor Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Resource-constrained computing devices, such as wearable devices, face challenges in processing large machine learning models due to limited computational power and battery life, leading to trade-offs in design features like form factor, size, and cost.
Innovation Solution
A multi-model system is implemented, comprising a gating model and a main model, where the gating model is trained to meet a minimum accuracy threshold with lower power consumption, allowing the system to switch between models based on battery life, thereby balancing resource consumption and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large machine learning model is processed to achieve high accuracy, then the model accuracy is improved, but the power consumption increases
Solution Approach 1:
The patent segments the machine learning processing into two distinct models: a gating model for initial classification and a main model for detailed processing. This segmentation allows the system to divide computational tasks, processing only critical inputs through the more resource-intensive main model while using the lighter gating model for routine inputs, thereby reducing overall power consumption while maintaining accuracy where needed.
Solution Approach 2:
The patent implements dynamic model selection based on input characteristics and device state. The system dynamically switches between the gating model and main model depending on factors such as battery life, input type, and processing requirements. This dynamic adaptation allows the system to optimize the balance between accuracy and power consumption in real-time, selecting the appropriate model for each specific processing scenario.
2Measurement precision
If the main model is always used to ensure high accuracy, then the model accuracy is maintained, but the device battery life is reduced
Solution Approach 1:
The patent applies partial action by using the gating model for a subset of inputs that do not require full processing. Instead of always executing the complete main model, the system performs only the necessary gating classification for routine inputs, reserving the more resource-intensive main model processing for cases where it is actually needed. This partial execution approach extends battery life while maintaining sufficient accuracy for most scenarios.
Solution Approach 2:
The patent changes the operational parameters of the machine learning system by switching between different models based on device state. When battery life is low or inputs suggest routine processing, the system parameters are adjusted to use the gating model instead of the main model. This parameter change allows the system to adapt its computational behavior to current conditions, extending battery life while maintaining acceptable performance levels.
3Power
If a resource-constrained device is designed with improved hardware to process larger models, then the processing capability is improved, but the device form factor and size increase
Solution Approach 1:
The patent creates a simplified copy of the main model in the form of the gating model. This copy retains the essential classification functionality but with reduced computational complexity, allowing it to run on resource-constrained devices without requiring enhanced hardware. The gating model serves as an efficient approximation that enables sophisticated processing capabilities to exist on compact devices.
Solution Approach 2:
By segmenting the processing into two models with different computational requirements, the patent enables resource-constrained devices to perform sophisticated machine learning tasks without needing upgraded hardware. The gating model handles routine processing on existing hardware, while the main model provides enhanced capability when needed, all within the same device form factor.
Data Source
AI summary
Methods, systems, and computer-readable media for multi-model processing on resource-constrained devices. A resource-constrained device can determine, based on a battery-life for a battery of the device, whether to process input through a first model or a second model. The first model can be a gating model that is more energy efficient to execute, and the second model can be a main model that is more accurate than the gating model. Depending on the current battery-life and/or other criteria, the system can process, through the gating model, sensor input that can record activity performed by a user of the resource-constrained device. If the gating model predicts an activity performed by the user that is recorded by the sensor data, the device can process the same or additional input through the main model. Overall power consumption can be reduced with a minimum accuracy maintained over processing input only through the main model.