Time-Series Vectorization via Frequency Domain Transformation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional vectorization approaches for time-series information represent data in the time domain, leading to large vectors that require significant memory, processing, and storage resources, and can over-represent data, limiting comparisons between time-series of varying lengths.

Innovation Solution

The method involves transforming time-series information into the frequency domain using techniques like Fast Fourier Transform (FFT), selecting a subset of lower frequency components, and performing a pairwise join to obtain a dimensionally-reduced frequency-domain representation, which is a one-dimensional vector of real numbers.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If conventional time-domain vectorization is used, then the complete time-series data is represented, but the vector size becomes large requiring significant memory and processing resources

Engineering Contradiction:

Improvetime-series data representation completenessVSAvoidvector size

Core Design Contradiction:

Loss of informationVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential frequency components from the complete frequency spectrum by applying a frequency threshold. This extraction process removes high-frequency components that are less important for capturing the general shape of time-series data, thereby reducing vector size while maintaining representation quality. The thresholding operation selectively retains only the most significant frequency components.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent uses approximate frequency domain representations instead of exact complete representations. By accepting a lossy approximation that captures only the essential characteristics (general shape) of the time-series data, the system achieves sufficient representation with much smaller vectors, trading exactness for efficiency.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

2Measurement precision

If complete frequency domain representation is used, then accurate time-series information is captured, but processing and storage resources increase

Engineering Contradiction:

Improvetime-series information accuracyVSAvoidcomputational resource consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies partial action by processing only a subset of frequency components rather than the complete spectrum. By focusing computational effort on only the lower frequency components that capture the essential characteristics of the data, the system achieves sufficient accuracy with reduced computational resources, avoiding the excessive processing required for complete frequency domain representation.

Inventive Principle:

Principle #16Partial or excessive action

3Loss of information

If time-domain vectorization is used, then all data points are included, but comparisons between time-series of varying lengths are limited

Engineering Contradiction:

Improvedata point inclusionVSAvoidcomparison capability across varying lengths

Core Design Contradiction:

Loss of informationVSAdaptability or versatility

Solution Approach 1:

The patent transforms time-series data from the time domain to the frequency domain, changing the dimensional perspective. In the frequency domain, each time-series is represented by its spectral components regardless of original length, enabling comparison between series of different durations. The frequency representation captures essential patterns that are invariant to time scaling, allowing versatile comparisons across varying lengths.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent changes the representation parameters from time-domain values to frequency-domain characteristics. By transforming the data into a different parameter space (frequency spectrum), the system enables comparison based on spectral similarity rather than temporal alignment, making comparisons between time-series of varying lengths feasible and meaningful.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20250147962A1Optimizing Vector Embedding Representations of Time-Series Information via Frequency Domain Representations

Publication Date: 2025.05.08 KX SYSTEMS INC

AI summary

A set of time-series information descriptive of one or more events occurring within a particular period of time is obtained. A frequency domain transformation is applied to the set of time-series information to obtain a frequency-domain representation of the set of time-series information comprising a plurality of frequency components. A dimensionally-reduced frequency-domain representation of the set of time-series information is determined based at least in part on a first subset of frequency components of the plurality of frequency components.