Time-Series Classification via Warping Distance Pruning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Time-series data generated from various devices, such as motion sensors and health sensors, often vary in speed, making it challenging to classify accurately, as existing methods struggle with speed differences and multi-dimensional data, leading to inefficient classification processes and high computational costs.
Innovation Solution
A real-time classification method that retrieves input data of a first window size and classifies it based on the comparison of warping distance with a pruning threshold, using a robust classification process that is invariant to speed differences, reducing costly warping distance computations through efficient lower bound computation and pruning schemes, enabling real-time processing of multi-dimensional time-series data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing classification methods are used for time-series data with different speeds, then classification can be performed, but accuracy deteriorates and computational cost increases
Solution Approach 1:
The system performs preliminary actions by pre-computing lower bound estimates of warping distances and pre-establishing pruning thresholds during training. This allows rapid classification during inference without performing costly full warping distance computations, thereby maintaining high accuracy while reducing computational energy consumption.
Solution Approach 2:
The invention extracts and utilizes only the essential information needed for classification by computing lower bound estimates rather than full warping distances. This selective extraction of critical features enables accurate classification with significantly reduced computational overhead, addressing the contradiction between precision and energy consumption.
2Measurement precision
If full warping distance computation is performed for all time-series data, then classification accuracy is maintained, but processing time increases
Solution Approach 1:
The system applies partial action by computing only the necessary lower bound estimates of warping distances rather than performing complete computations. This partial computation approach is sufficient for accurate classification when combined with the pruning threshold, significantly reducing processing time while maintaining precision.
Solution Approach 2:
The invention enables skipping unnecessary computational steps by using lower bound estimates to quickly identify and discard non-matching time-series data. This rushing through of the classification process for obvious non-matches reduces overall processing time while maintaining accuracy through the pruning threshold mechanism.
3Speed
If classification is performed in real-time for multi-dimensional time-series data, then responsiveness is improved, but computational complexity increases
Solution Approach 1:
The system reduces real-time computational complexity by performing all heavy lifting during the training phase through pre-computation of lower bound estimates and pruning thresholds. During real-time inference, only simple comparisons are needed, enabling fast processing while keeping device complexity manageable.
Solution Approach 2:
The invention segments the classification task into two distinct phases: a training phase where complex computations are performed to establish pruning thresholds and lower bound estimates, and an inference phase where simple comparisons are made for real-time classification. This segmentation enables real-time processing by separating computational complexity into appropriate temporal phases.
Data Source
AI summary
A recognition method includes retrieving an input including data of a first window size. The method further includes classifying the input based on comparison of warping distance of the input with a pruning threshold.