Topological Attention for Time Series Forecasting

Sebastian Zeng, Florian Graf, Christoph Hofer, Roland Kwitt

Keywords: Time series forecasting, Persistent homology, Attention, Topological Data Analysis

TL;DR: We study whether local topological properties of time series, as captured via persistent homology, can serve as a reliable signal that provides complementary information for learning to forecast.

Abstract: The problem of (point) forecasting univariate time series is considered. Most approaches, ranging from traditional statistical methods to recent learning-based techniques with neural networks, directly operate on raw time series observations. As an extension, we study whether local topological properties, as captured via persistent homology, can serve as a reliable signal that provides complementary information for learning to forecast. To this end, we propose topological attention, which allows attending to local topological features within a time horizon of historical data. Our approach easily integrates into existing end-to-end trainable forecasting models, such as N-BEATS, and, in combination with the latter exhibits state-of-the-art performance on the large-scale M4 benchmark dataset of 100,000 diverse time series from different domains. Ablation experiments, as well as a comparison to recent techniques in a setting where only a single time series is available for training, corroborate the beneficial nature of including local topological information through an attention mechanism.

Code Of Conduct: I certify that all co-authors of this work have read and commit to adhering to the NeurIPS Statement on Ethics, Fairness, Inclusivity, and Code of Conduct.

Code: https://github.com/plus-rkwitt/TAN

Community Implementations: [ 4 code implementations](https://www.catalyzex.com/paper/arxiv:2107.09031/code)