Advancements in Time Series Forecasting

The field of time series forecasting is experiencing significant growth, with a focus on improving the accuracy and efficiency of forecasting models. Recent developments have centered around incorporating external information, such as exogenous inputs, to enhance model performance. Additionally, there is a growing interest in decomposing the time series forecasting pipeline into modular components, allowing for more effective sequence representation, information extraction, and target projection. Other notable trends include the use of deep learning architectures, such as transformers and convolutional layers, to extract spatial-temporal dependencies and model complex patterns in time series data. Furthermore, researchers are exploring the development of foundation models that can generalize across diverse prediction tasks and scenarios, as well as techniques for interpreting and explaining the behavior of time series models. Noteworthy papers include Temporal Window Smoothing of Exogenous Variables for Improved Time Series Prediction, which proposes a method for whitening exogenous inputs to reduce redundancy and improve forecasting performance. Another significant contribution is the paper Decomposing the Time Series Forecasting Pipeline, which achieves state-of-the-art forecasting accuracy while enhancing computational efficiency. The paper MoFE-Time also stands out for its innovative approach to integrating time and frequency domain features within a Mixture of Experts network, achieving new state-of-the-art performance on several benchmarks.

Sources

Temporal Window Smoothing of Exogenous Variables for Improved Time Series Prediction

EmissionNet: Air Quality Pollution Forecasting for Agriculture

Decomposing the Time Series Forecasting Pipeline: A Modular Approach for Time Series Representation, Information Extraction, and Projection

A Wireless Foundation Model for Multi-Task Prediction

KnowIt: Deep Time Series Modeling and Interpretation

MoFE-Time: Mixture of Frequency Domain Experts for Time-Series Forecasting Models

Bridging the Last Mile of Prediction: Enhancing Time Series Forecasting with Conditional Guided Flow Matching

Towards Interpretable Time Series Foundation Models

Built with on top of