Multi-Scale Time Series Causal Analysis With Hierarchical Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional causal analysis systems fail to determine causal relationships related to underlying temporal phenomena associated with time series data due to analyzing at fixed time scales, ignoring temporal dynamics.
Innovation Solution
A computing system converts large-scale time series data into records at multiple time resolutions and uses hierarchical transformer models to generate graph data structures that capture temporal dynamics, enabling efficient causal analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If causal analysis is performed at fixed time scales, then the analysis process is simple, but temporal dynamics of time series data are not captured
Solution Approach 1:
The patent segments the time series data into multiple time resolutions (e.g., hourly, daily, weekly aggregates) to capture temporal dynamics at different scales. This segmentation allows the system to analyze causal relationships while preserving temporal information that would be lost in fixed-scale analysis.
Solution Approach 2:
The patent adds a time resolution dimension to the causal analysis framework. Instead of analyzing data at a single fixed time scale, the system incorporates multiple time resolutions as an additional dimension, enabling capture of temporal dynamics while maintaining analytical tractability.
2Measurement precision
If multiple time resolutions are used to capture temporal dynamics, then causal patterns are improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary aggregation of time series data into multiple time resolutions before conducting causal analysis. By pre-processing the data into hierarchical time aggregates, the system reduces the computational burden of analyzing raw high-frequency data while preserving temporal dynamics information.
Solution Approach 2:
The patent implements a dynamic causal analysis framework that adapts to different time resolutions. The system dynamically selects and combines results from multiple time resolution analyses, allowing it to capture temporal dynamics efficiently without requiring exhaustive computation at all possible resolutions simultaneously.
3Measurement precision
If temporal dynamics are captured through hierarchical models, then causal analysis quality improves, but processing time increases
Solution Approach 1:
The patent segments the hierarchical processing into distinct levels (e.g., fine-grained hourly patterns, medium-grained daily patterns, coarse-grained weekly patterns). Each level processes data independently at its appropriate resolution, avoiding the computational overhead of processing all data at the finest resolution while maintaining accuracy of temporal characteristics.
Solution Approach 2:
The patent applies partial processing by focusing computational resources on the most relevant time resolutions for each specific causal analysis task. Rather than exhaustively processing all possible time resolutions, the system selectively processes only those resolutions necessary to capture the temporal dynamics of interest, reducing overall processing time.
Data Source
AI summary
In general, various aspects of the techniques are directed to causal analysis using large scale time series data. A computing system may convert large scale time series data to first time period records and second time period records according to a multi-scale time resolution. The computing system may implement a hierarchical machine learning model to generate embeddings that capture temporal characteristics of features of the large scale time series data. The computing system may generate a graph data structure indicating cause and effect correlations between features of the large scale time series data based on temporal dynamics captured in the cause and second time period records and/or the embeddings.