Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Task-Synchronized Recurrent Neural Networks

Mantas Lukoševičius, Arnas Uselis

TL;DR

This work tackles irregular time sampling in time series by reframing RNNs as continuous-time dynamical systems and resampling them in time rather than resampling the data. It introduces Task-Synchronized ESN (TSESN) and Task-Synchronized GRU (TSGRU), which update their hidden states using data-driven time steps $\Delta t_n$ while preserving standard readouts, requiring no data interpolation. The approach is analytically grounded and empirically validated on synthetic chaotic datasets and real-world tasks (gesture recognition and speleothem time series), showing robustness to time irregularities and competitive performance against resampling and other time-aware models. The authors provide open-source code, highlighting practical benefits in speed and simplicity for handling irregular sampling in RNNs.

Abstract

Data are often sampled irregularly in time. Dealing with this using Recurrent Neural Networks (RNNs) traditionally involved ignoring the fact, feeding the time differences as additional inputs, or resampling the data. All these methods have their shortcomings. We propose an elegant straightforward alternative approach where instead the RNN is in effect resampled in time to match the time of the data or the task at hand. We use Echo State Network (ESN) and Gated Recurrent Unit (GRU) as the basis for our solution. Such RNNs can be seen as discretizations of continuous-time dynamical systems, which gives a solid theoretical ground to our approach. Our Task-Synchronized ESN (TSESN) and GRU (TSGRU) models allow for a direct model time setting and require no additional training, parameter tuning, or computation (solving differential equations or interpolating data) compared to their regular counterparts, thus retaining their original efficiency. We confirm empirically that our models can effectively compensate for the time-non-uniformity of the data and demonstrate that they compare favorably to data resampling, classical RNN methods, and alternative RNN models proposed to deal with time irregularities on several real-world nonuniform-time datasets. We open-source the code at https://github.com/oshapio/task-synchronized-RNNs .

Task-Synchronized Recurrent Neural Networks

TL;DR

This work tackles irregular time sampling in time series by reframing RNNs as continuous-time dynamical systems and resampling them in time rather than resampling the data. It introduces Task-Synchronized ESN (TSESN) and Task-Synchronized GRU (TSGRU), which update their hidden states using data-driven time steps while preserving standard readouts, requiring no data interpolation. The approach is analytically grounded and empirically validated on synthetic chaotic datasets and real-world tasks (gesture recognition and speleothem time series), showing robustness to time irregularities and competitive performance against resampling and other time-aware models. The authors provide open-source code, highlighting practical benefits in speed and simplicity for handling irregular sampling in RNNs.

Abstract

Data are often sampled irregularly in time. Dealing with this using Recurrent Neural Networks (RNNs) traditionally involved ignoring the fact, feeding the time differences as additional inputs, or resampling the data. All these methods have their shortcomings. We propose an elegant straightforward alternative approach where instead the RNN is in effect resampled in time to match the time of the data or the task at hand. We use Echo State Network (ESN) and Gated Recurrent Unit (GRU) as the basis for our solution. Such RNNs can be seen as discretizations of continuous-time dynamical systems, which gives a solid theoretical ground to our approach. Our Task-Synchronized ESN (TSESN) and GRU (TSGRU) models allow for a direct model time setting and require no additional training, parameter tuning, or computation (solving differential equations or interpolating data) compared to their regular counterparts, thus retaining their original efficiency. We confirm empirically that our models can effectively compensate for the time-non-uniformity of the data and demonstrate that they compare favorably to data resampling, classical RNN methods, and alternative RNN models proposed to deal with time irregularities on several real-world nonuniform-time datasets. We open-source the code at https://github.com/oshapio/task-synchronized-RNNs .
Paper Structure (15 sections, 10 equations, 2 figures, 2 tables)

This paper contains 15 sections, 10 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Direct Time to Output Mapping
  4. RNN Models with Data Pre-Treatment Layers
  5. RNN Models with Special Gates
  6. RNN Methods Employing Differential Equation Solvers
  7. Methods
  8. Task-Synchronized Echo State Networks
  9. Task-Synchronized Gated Recurrent Units
  10. Nonlinear Time Scaling
  11. Experiments
  12. Synthetic Chaotic Attractor Datasets
  13. UWave Gesture Dataset
  14. Speleothem Dataset
  15. Discussion

Figures (2)

  • Figure 1: Comparison of ESN, ESN+$\Delta t$, and TSESN mean error for varying time irregularity of chaotic attractor data. Shading indicates min/max errors over 50 runs.
  • Figure 2: Time and performance comparison on UWaveGesture dataset. Proposed variations are marked in orange. Simple ESN was omitted due to subpar performance on this task.