Anchor Window Selection in Time Series Representation Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional time series representation learning methods using dilated temporal convolutional neural networks with triplet loss suffer from wide variations in classification accuracy across training runs due to random selection of anchor window sizes and positions, which neglects the natural measures of similarity and contrast in numerical time series data.
Innovation Solution
A computing device method to determine anchor window size and position in triplet loss formulation by calculating moving averages and standard deviations, iteratively expanding anchors to cover visited neighbors, and setting window sizes based on expanded anchors, thereby improving representation learning accuracy and reducing variations in classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If random anchor window selection is used, then assembly of triplets is easy, but classification accuracy shows wide variations across training runs
Solution Approach 1:
The patent changes the parameters of anchor window selection from random to systematic based on statistical measures. It calculates moving averages and standard deviations to identify meaningful anchor positions, transforming the selection process from stochastic to deterministic based on data characteristics. This resolves the contradiction by maintaining ease of implementation while significantly improving accuracy consistency across training runs.
Solution Approach 2:
The patent enables the data itself to guide the anchor window selection process through self-service mechanisms. By computing moving averages and standard deviations from the time series data, the system automatically identifies informative anchor positions without external intervention. This self-organizing approach eliminates random variation while preserving implementation simplicity.
2Device complexity
If conventional random triplet selection is used, then implementation is simple, but representation learning accuracy is limited
Solution Approach 1:
The patent applies preliminary action by pre-processing the time series data to calculate moving averages and standard deviations before triplet construction. This preparatory step identifies informative regions and guides subsequent anchor selection, improving representation learning accuracy without significantly increasing implementation complexity. The pre-computed statistical measures serve as a foundation for more accurate triplet assembly.
3Productivity
If fixed window size is used, then processing is efficient, but optimization converges slowly and gets trapped in local minima
Solution Approach 1:
The patent introduces dynamics by making the anchor window size adaptive rather than fixed. The window size is determined by the calculated standard deviation and moving average characteristics of the data, allowing it to adjust to different data patterns. This dynamic approach improves optimization convergence speed and reduces the risk of getting trapped in local minima while maintaining reasonable processing efficiency through systematic selection.
Data Source
AI summary
A method of using a computing device to determine a window size in variate time series data that includes receiving, by a computing device, variate time series data associated with a machine learning model. The computing device sets a moving window size and a standard deviation for the variate time series data. The computing device further calculates a moving window average for the variate time series data. The computing device additionally calculates a standard deviation across all variate time series data. The computing device sorts the standard deviations calculated in descending order. The computing device further iterates indices for the standard deviations until the indices have been visited by at least one anchor. The computing device iteratively expands each anchor to cover neighbors' anchors which have been visited by previous anchors. The computing device determines a window size based upon the expanded anchors.