Recent advancements in time series forecasting are increasingly focused on enhancing model efficiency and accuracy while addressing the complexities of real-world data. New frameworks are emerging that prioritize long-term forecasting stability, such as those that utilize look-ahead augmentation and self-corrective refinement to improve predictive consistency. Additionally, linear models are gaining traction for their ability to reduce computational complexity, enabling faster inference without sacrificing performance. The introduction of global temporal retrieval mechanisms allows models to capture long-range dependencies more effectively, while dual-path architectures are being explored to separate autoregressive dynamics from cross-dimensional interactions, enhancing overall accuracy. Furthermore, the integration of mixture-of-experts approaches is proving effective in managing diverse temporal patterns, improving robustness against noise, and facilitating uncertainty quantification. These developments suggest a shift towards more interpretable and efficient forecasting solutions that can better serve industries reliant on accurate predictions, such as finance, healthcare, and education.
Top papers
- Back to the Future: Look-ahead Augmentation and Parallel Self-Refinement for Time Series Forecasting(8.0)
- vLinear: A Powerful Linear Model for Multivariate Time Series Forecasting(7.0)
- Seg-MoE: Multi-Resolution Segment-wise Mixture-of-Experts for Time Series Forecasting Transformers(7.0)
- SEA-TS: Self-Evolving Agent for Autonomous Code Generation of Time Series Forecasting Algorithms(7.0)
- Enhancing Multivariate Time Series Forecasting with Global Temporal Retrieval(7.0)
- AltTS: A Dual-Path Framework with Alternating Optimization for Multivariate Time Series Forecasting(7.0)
- DualWeaver: Synergistic Feature Weaving Surrogates for Multivariate Forecasting with Univariate Time Series Foundation Models(7.0)
- CAPS: Unifying Attention, Recurrence, and Alignment in Transformer-based Time Series Forecasting(5.0)
- StretchTime: Adaptive Time Series Forecasting via Symplectic Attention(5.0)
- PatchDecomp: Interpretable Patch-Based Time Series Forecasting(5.0)
- Let Experts Feel Uncertainty: A Multi-Expert Label Distribution Approach to Probabilistic Time Series Forecasting(5.0)
- Commencing-Student Enrolment Forecasting Under Data Sparsity with Time Series Foundation Models(5.0)
- SDMixer: Sparse Dual-Mixer for Time Series Forecasting(5.0)
- DropoutTS: Sample-Adaptive Dropout for Robust Time Series Forecasting(5.0)
- MoHETS: Long-term Time Series Forecasting with Mixture-of-Heterogeneous-Experts(5.0)
- Reverso: Efficient Time Series Foundation Models for Zero-shot Forecasting(2.0)