The field of time series forecasting is moving towards leveraging complex spatio-temporal dynamics and multivariate long-term history representation to improve prediction accuracy. Recent developments have focused on effectively capturing spatial-temporal traffic dynamics, incorporating cross-future behavior, and modeling inter-series dependencies. Notable advancements include the use of time-aware transformers, graph learning modules, and mixture-of-experts-enhanced foundation models. These innovations have led to significant improvements in predicting accuracy, model generalization, and adaptability. Noteworthy papers include: Learning Spatio-Temporal Dynamics for Trajectory Recovery via Time-Aware Transformer, which proposes a novel method for trajectory recovery using time-aware transformers. CRAFT: Time Series Forecasting with Cross-Future Behavior Awareness, which explores the use of cross-future behavior to mine trends in time series data. Leveraging Multivariate Long-Term History Representation for Time Series Forecasting, which proposes a framework for incorporating long-term history into modeling. Time Tracker: Mixture-of-Experts-Enhanced Foundation Time Series Forecasting Model with Decoupled Training Pipelines, which leverages sparse mixture of experts and graph learning modules to handle diverse time series patterns.