Training Data Sampling Device for Time-Series Temporal Feature Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The sliding window method for sampling training data from time-series data often misses temporal features, leading to inefficient data selection and reduced accuracy in detecting abrupt changes and segmenting rapidly changing labels, which can result in overfitting and poor performance of artificial neural networks.
Innovation Solution
A training data sampling device and method that estimates label intervals using semantic segmentation and change point detection algorithms to identify temporal features, determining a sampling size based on the maximum separation distance between change points, thereby sampling training data that includes significant events in time-series data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If sliding window method is used to sample training data, then the sampling process is simple and fast, but temporal features are missed and segmentation accuracy is reduced
Solution Approach 1:
The patent applies preliminary action by performing change point detection and label interval estimation before the actual sampling process. The controller identifies temporal features and determines optimal sampling points in advance, ensuring that training data includes significant events while maintaining efficient sampling.
2Ease of operation
If fixed window size with fixed stride is used, then the sampling process is straightforward, but the temporal feature may be completely omitted from time-series data
Solution Approach 1:
The patent implements dynamics by making the window size and sampling points adaptive rather than fixed. The controller dynamically adjusts sampling parameters based on detected change points and label intervals, allowing the sampling process to adapt to the actual temporal characteristics of the time-series data while maintaining operational simplicity.
3Productivity
If sliding window method samples without preprocessing, then the process is efficient, but unnecessary training data is sampled and temporal features are not accurately detected
Solution Approach 1:
The patent applies the taking out principle by extracting and focusing on only the necessary portions of time-series data. The controller identifies change points and label intervals to extract relevant temporal features, then samples training data specifically around these identified points, eliminating unnecessary data while preserving critical information.
Data Source
AI summary
A training data sampling device and a training data sampling method may improve the reliability of training data by estimating at least one label interval for time-series data, detecting a temporal feature from the time-series data, and sampling the training data including the temporal feature for the respective label interval. To the end, the training data sampling device may include an input device that receives time-series data and may include a controller that estimates at least one label interval from the time-series data, detects a temporal feature from the time-series data, and samples training data including the temporal feature by the label interval.