Cascaded Machine Learning System for Wearable Power Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine-learning algorithms for classifying input data from inertial sensors are excessively large, requiring significant computational power and storage resources, making them unsuitable for deployment in resource-constrained settings like wearable devices and IoT devices.
Innovation Solution
A cascaded machine-learning system comprising two subsystems, where a first subsystem with a less complex machine-learning algorithm classifies data initially, and a second subsystem with a more complex algorithm is activated only when necessary, conserving power and reducing overall energy consumption by splitting functionality across processors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single large machine-learning algorithm is used for classification, then classification accuracy is improved, but computational power requirements and energy consumption increase significantly
Solution Approach 1:
The patent divides the classification system into two subsystems: a first subsystem with a less complex machine-learning algorithm for initial classification, and a second subsystem with a more complex algorithm for refined classification. This segmentation allows the system to achieve high classification accuracy when needed while reducing overall energy consumption by using the simpler algorithm for routine classifications.
Solution Approach 2:
The patent implements dynamic activation of subsystems based on operating conditions. The second subsystem is activated only when necessary (e.g., when the first subsystem's confidence level is below a threshold or when invalid classifications are detected), while the first subsystem operates normally for most cases. This dynamic behavior optimizes the trade-off between accuracy and energy consumption.
2Measurement precision
If a single large machine-learning algorithm is deployed, then classification performance is improved, but device complexity and resource requirements increase
Solution Approach 1:
The patent segments the computational workload into two parts by deploying two subsystems with different algorithm complexities. The first subsystem handles routine classifications with minimal resources, while the second subsystem provides enhanced performance only when required, thereby reducing overall device complexity compared to deploying a single large algorithm continuously.
Solution Approach 2:
The system applies partial action by using the simpler first subsystem for most classification tasks and only invoking the more complex second subsystem when necessary (e.g., for edge cases, low confidence predictions, or invalid classifications). This partial deployment of computational resources achieves high performance when needed while maintaining lower overall complexity.
3Reliability
If the second subsystem is always active, then classification reliability is improved, but power consumption increases
Solution Approach 1:
The patent implements dynamic activation where the second subsystem is activated only when the first subsystem detects conditions requiring enhanced classification (e.g., low confidence levels, invalid classifications, or specific operating conditions). This dynamic behavior maintains high reliability for critical classifications while significantly reducing power consumption compared to continuous operation of both subsystems.
Solution Approach 2:
The system uses feedback mechanisms where the first subsystem monitors its own output (confidence levels, classification validity) and triggers activation of the second subsystem based on this feedback. This feedback-driven activation ensures reliability is maintained for problematic cases while avoiding unnecessary power consumption from continuous second subsystem operation.
Data Source
AI summary
A cascaded machine learning inference system and method is disclosed. The cascaded system and method may be designed to be employed in resource restricted environments. The cascaded system and method may be applicable for applications that operate with limited power (e.g., a wearable smart watch). The cascaded system and method may employ two or more subsystems that are operable to classify an input signal provided by any number or types of sensors suitable for a given application. For instance, the sensors used may include gyroscopes, accelerometers, magnetometers, or barometric altimeters. The system and method may also be further split functionality across additional or new subsystems. By splitting operations and functionality across additional subsystems, the overall power consumption may further be reduced.