The field of time series analysis is moving towards a greater emphasis on causal modeling and the development of more advanced techniques for forecasting and anomaly detection. Researchers are exploring new methods for identifying causal relationships between variables, such as the use of Multivariate Granger Causality and PCMCI+, and for incorporating this information into predictive models. Additionally, there is a growing interest in the use of foundation models and other deep learning architectures for time series forecasting and anomaly detection. These models have been shown to be highly effective in a variety of applications, including the prediction of Arctic sea ice extent and the detection of anomalies in time series data. Notable papers in this area include: Learning What Matters: Causal Time Series Modeling for Arctic Sea Ice Prediction, which introduces a causality-aware deep learning framework for predicting Arctic sea ice extent. C3DE: Causal-Aware Collaborative Neural Controlled Differential Equation for Long-Term Urban Crowd Flow Prediction, which proposes a novel framework for modeling the dynamics of urban crowd flow. Leveraging Intermediate Representations of Time Series Foundation Models for Anomaly Detection, which presents a new approach to anomaly detection using the intermediate representations of time series foundation models.
Causal Modeling and Advanced Time Series Analysis
Sources
Learning What Matters: Causal Time Series Modeling for Arctic Sea Ice Prediction
ARMA Block: A CNN-Based Autoregressive and Moving Average Module for Long-Term Time Series Forecasting
Multiscaling in Wasserstein Spaces
C3DE: Causal-Aware Collaborative Neural Controlled Differential Equation for Long-Term Urban Crowd Flow Prediction
Prediction and Causality of functional MRI and synthetic signal using a Zero-Shot Time-Series Foundation Model
Cross-Modal Deep Metric Learning for Time Series Anomaly Detection
Leveraging Intermediate Representations of Time Series Foundation Models for Anomaly Detection
TimeCluster with PCA is Equivalent to Subspace Identification of Linear Dynamical Systems
Forward Euler for Wasserstein Gradient Flows: Breakdown and Regularization
DyWPE: Signal-Aware Dynamic Wavelet Positional Encoding for Time Series Transformers
DeCoP: Enhancing Self-Supervised Time Series Representation with Dependency Controlled Pre-training
DPANet: Dual Pyramid Attention Network for Multivariate Time Series Forecasting
DAG: A Dual Causal Network for Time Series Forecasting with Exogenous Variables
Warp Quantification Analysis: A Framework For Path-based Signal Alignment Metrics
AnoF-Diff: One-Step Diffusion-Based Anomaly Detection for Forceful Tool Use