Autoregressive Convolutional Neural Networks for Asynchronous Time Series

By Mikolaj Binkowski, Gautier Marti, Philippe Donnat. 2017

164 citations recorded by the source corpus.

We propose Significance-Offset Convolutional Neural Network, a deep convolutional network architecture for regression of multivariate asynchronous time series. The model is inspired by standard autoregressive (AR) models and gating mechanisms used in recurrent neural networks. It involves an AR-like weighting system, where the final predictor is obtained as a weighted sum of adjusted regressors, while the weights are datadependent functions learnt through a convolutional network. The architecture was designed for applications on asynchronous time series and is evaluated on such datasets: a hedge fund proprietary dataset of over 2 million quotes for a credit derivative index, an artificially generated noisy autoregressive series and UCI household electricity consumption dataset. The proposed architecture achieves promising results as compared to convolutional and recurrent neural networks.

Subfields: Electrical and Electronic Engineering, Management Science and Operations Research, Model Architectures, Signal Processing, Time-Series & Forecasting

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Electrical and Electronic Engineering

In Management Science and Operations Research

In Model Architectures

In Signal Processing

In Time-Series & Forecasting

Citations in the corpus

Enhancing Time-Series Momentum Strategies Using Deep Neural Networks (2019, methodology): “[19].”
Conditional Time Series Forecasting with Convolutional Neural Networks (2017, methodology): “This problem of the imbalance between the need of memory and the non-linearities was also adressed in [3] by using a combination of an autoregressive model and a CNN.”
Dilated convolutional neural networks for time series forecasting (2018, methodology): “This problem of the imbalance between the need of memory and the non-linearities was also adressed in [3] by using a combination of an autoregressive model and a CNN.”
Practical Approach to Asynchronous Multivariate Time Series Anomaly Detection and Localization (2021, background): “In a more recent study on noisy multivariate forecast problems [5], the authors tackled asynchronous dimensions by merging them into a single dimension while adding their asynchronous temporal attributes into separate features.”
Variable-rate discrete representation learning (2021, background): “Many such variants of recurrent and convolutional neural networks have been explored (Mozer et al., 2017; Neil et al., 2017; Binkowski et al., 2018; Wu et al., 2018b; Rubanova et al., 2019; Shukla & Marlin, 2019).”
Day-ahead building-level load forecasts using deep learning vs. traditional time-series techniques (2019, methodology): “Authors in [33] innovatively created an auto-regressive-like double-CNN model for asynchronous time-series fore-casts.”
Causal Discovery with Attention-Based Convolutional Neural Networks (2019, methodology): “Convolutional architectures for time series are still scarce, but deep convolutional architectures were recently used for noisy financial time series forecasting [40] and for multivariate asynchronous time series prediction [41].”