Multivariate Time Series Clustering via Embedding Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in effectively clustering multivariate discrete time series data to understand disease progression patterns and design tailored treatment guidelines, due to difficulties in discovering predictive patterns across temporal correlations and estimating label distributions.
Innovation Solution
A method and apparatus for data clustering using a first artificial neural network to generate embedding vectors for multivariate discrete time series data, followed by a path-based connectivity test to create a similarity graph, and a second artificial neural network to predict label distributions, allowing for effective clustering based on these graphs and distributions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional clustering methods are used on multivariate discrete time series data, then the process is simpler, but the ability to discover predictive patterns across temporal correlations is insufficient
Solution Approach 1:
The patent introduces embedding vectors as an intermediary representation that transforms complex multivariate discrete time series data into a compressed latent space. This embedding layer mediates between the raw data and the clustering algorithm, enabling pattern discovery while managing complexity through dimensionality reduction and temporal correlation capture.
Solution Approach 2:
The patent segments the clustering process into distinct functional modules: data encoding to embedding vectors, similarity graph construction from embedding vectors, and label distribution prediction. This segmentation allows each component to be optimized independently while collectively solving the pattern discovery problem.
2Measurement precision
If complex path-based connectivity tests are performed to create similarity graphs, then clustering accuracy improves, but computational time increases
Solution Approach 1:
The patent performs preliminary encoding of all time series data into embedding vectors before constructing the similarity graph. This preliminary action pre-processes the data into a compact representation that captures temporal correlations, making subsequent connectivity tests more efficient while maintaining high clustering accuracy.
Solution Approach 2:
The patent transforms the data from its original high-dimensional time series form into embedding vectors with different dimensional parameters. This parameter change reduces the computational complexity of connectivity tests while preserving the essential temporal correlation structures needed for accurate clustering.
3Measurement precision
If multiple artificial neural networks are used for embedding and label prediction, then the ability to characterize disease progression improves, but the system becomes more complex
Solution Approach 1:
The patent designs the neural network architecture where embedding vectors serve multiple functions: they represent temporal correlations, enable similarity computation, and serve as input for label distribution prediction. This multi-functionality reduces the need for separate specialized networks while maintaining high disease progression characterization accuracy.
Solution Approach 2:
The embedding vectors act as an intermediary that bridges the raw time series data and the label distribution prediction. This intermediary layer captures complex temporal patterns while keeping the prediction network simpler, as it receives pre-processed embeddings rather than raw multivariate data.
Data Source
AI summary
A method for data clustering performed in a computing device including one or more processors and a memory that stores one or more programs executed by the one or more processors according to an embodiment includes receiving two or more multivariate discrete time series data as input and generating an embedding vector for each of the multivariate discrete time series data using a first artificial neural network, generating a similarity graph by performing a path-based connectivity test between embedding vectors for each of the two or more multivariate discrete time series data, predicting a label distribution for embedding vectors for each of the two or more multivariate discrete time series data using a second artificial neural network, and clustering the two or more multivariate discrete time series data based on the similarity graph and the label distribution.