Time-Aligned Reconstruction RNN for Irregular Multivariate Time Series
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Recurrent neural networks (RNNs) face challenges in handling multivariate time series data with irregular time intervals and missing values, as existing methods are ineffective in addressing these issues, particularly in applications like electronic health records where time intervals between visits vary and not all variables are observed.
Innovation Solution
A time-aligned reconstruction recurrent neural network (TR-RNN) architecture that imputes missing values using a weighted mean and transforms inputs to uniformly time-aligned representations, incorporating time intervals to treat irregular time series data as regular time series data, thereby eliminating the need for the RNN to consider irregular time intervals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing RNN methods are used to handle missing values, then missing values can be addressed, but irregular time intervals cannot be handled effectively
Solution Approach 1:
The patent separates the handling of missing values and irregular time intervals into distinct modules: an imputation network for missing values and a time-aligned reconstruction network for irregular time intervals. This segmentation allows each module to specialize in one aspect, resolving the contradiction between handling missing values effectively and adapting to irregular time intervals.
Solution Approach 2:
The patent introduces time interval data as an intermediary element that mediates between the input data and the RNN processing. By incorporating time interval information into the reconstruction process, the system can handle irregular time intervals while maintaining the ability to process missing values through the imputation module.
2Ease of operation
If regular time series data is used as input to RNN, then the RNN can process data effectively, but irregular time intervals are not preserved
Solution Approach 1:
The patent adds a time interval dimension to the traditional time series data by incorporating time interval data into the reconstruction process. This dimensional extension allows the RNN to process data in a regularized format while preserving the temporal spacing information through the time-aligned reconstruction mechanism.
Solution Approach 2:
The patent transforms the time series data by applying parameter changes that incorporate time interval information. The reconstruction network learns to adjust the data parameters based on time interval patterns, enabling the RNN to process regularized data while the learned parameters preserve the original irregular time interval characteristics.
3Adaptability or versatility
If time interval information is incorporated into the model, then irregular time intervals can be handled, but model complexity increases
Solution Approach 1:
The patent merges the time interval processing functionality into the existing RNN architecture through the time-aligned reconstruction module. Rather than adding completely separate complex mechanisms, the time interval information is integrated into the reconstruction process, combining multiple functions into a unified model structure that handles both missing values and irregular time intervals.
Data Source
AI summary
A computer-implemented method for reconstructing time series data including irregular time intervals and missing values to predict future data from the time series data using a Recurrent Neural Network (RNN) is provided including obtaining irregular time series data X={x1, . . . , xt, . . . , xT} and time interval data Δ={δ1, . . . , δt, . . . , δT}, where xt is a D-dimensional feature vector, T is a total number of observations, δt is a D-dimensional time interval vector, and a d-th element δtd of δt represents a time interval from a last observation, replacing missing values in xt with imputed values using an imputation to obtain {tilde over (x)}t, rescaling data of the time interval δt to obtain rescaled time interval data φ(δt) by calculating φ(δt)=φ log(e+ max(0,ϕδt+bϕ))+bφ, where Wφ, Wϕ, bϕ, bφ are network parameters of a neural network and e is Napier's constant, and multiplying {tilde over (x)}t by φ(δt) to obtain {circumflex over (x)}t as regular time series data for input of the RNN.