Time Series Data Subsampling for Wearable Device Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional machine learning approaches for analyzing time-series data are inadequate in identifying individual localized points of interest, often missing critical data points and requiring entire regions, which can lead to incorrect interpretations and processing delays, especially in resource-constrained devices like wearable devices.
Innovation Solution
An adaptive framework that filters data to reduce processing requirements, allowing for real-time identification of localized points of interest by using sub sampling to select candidate samples based on mathematical properties, and classifying them using a model adapted to the changing characteristics of the data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning approaches are used to analyze time-series data, then the analysis can identify regions of interest, but it misses individual localized data points and introduces processing delays
Solution Approach 1:
The patent segments the time-series data analysis into two distinct phases: (1) a filtering phase that identifies candidate regions using traditional machine learning, and (2) a precise identification phase that examines individual localized data points within those regions. This segmentation allows the system to maintain both efficiency in region identification and precision in point-level detection, resolving the contradiction between identification accuracy and processing delay
Solution Approach 2:
The patent applies preliminary filtering actions to narrow down the search space before performing detailed analysis. By first identifying candidate regions using traditional machine learning approaches, then focusing subsequent computational resources only on those specific regions for localized point identification, the system reduces overall processing time while maintaining high accuracy in detecting individual data points
2Device complexity
If traditional machine learning approaches require entire regions to be identified, then processing can be simplified, but critical individual data points may be missed and interpretations may be incorrect
Solution Approach 1:
The patent divides the analysis into hierarchical levels: region-level filtering followed by point-level identification. This segmentation enables the system to maintain simple processing at the region level while applying more complex, accurate analysis only where necessary at the individual data point level, thus preserving both low overall complexity and high reliability
Solution Approach 2:
The patent applies different analysis qualities to different parts of the data: traditional machine learning methods are used for region-level filtering where simpler processing suffices, while more sophisticated localized analysis is applied only to specific data points within those regions where high precision is critical. This local differentiation of quality maintains simplicity where possible while ensuring accuracy where required
3Ease of operation
If resource-constrained devices like wearable devices are used, then portability and accessibility are improved, but computational resources are limited making real-time analysis difficult
Solution Approach 1:
The patent segments the computational workload into two phases: a lightweight filtering phase that can run efficiently on resource-constrained devices, and a more intensive precise identification phase that focuses only on candidate regions. This segmentation enables real-time processing on wearable devices by ensuring that computationally expensive operations are performed only on narrowed-down candidate data rather than the entire time-series dataset
Solution Approach 2:
The patent applies partial action by performing complete detailed analysis only on a subset of candidate regions identified by the filtering phase, rather than analyzing the entire time-series data with high precision. This approach enables real-time processing on resource-constrained devices by applying excessive computational resources only where necessary while using simpler methods elsewhere
Data Source
AI summary
Methods, apparatus, systems and articles of manufacture to analyze time series data are disclosed. An example method includes sub sampling time series data collected by a sensor to generate one or more candidate samples of interest within the time series data. Feature vectors are generated for respective ones of the one or more candidate samples of interest. Classification of the feature vectors is attempted based on a model. In response to a classification of one of the feature vectors, the classification is stored in connection with the corresponding candidate sample.