arXiv：cs.LG· Jung Min Choi, Ngoc Son Le, Ibram Abdelmalak, Vijaya Krishna Yalavarthi, Lars Schmidt-Thieme·· 3 小时前AI 评分32
MARO：面向时间序列预测的最近锚定与循环排序模型
Most-Recent Anchoring with Recurrent Ordering for Time Series Forecasting
AI 导读
研究者提出 MARO，一种从最近 patch 到最旧 patch 处理回看窗口的时间序列预测模型，以最近 patch 作为锚点初始化隐状态并条件化后续每一步，使旧 patch 被折叠进以近期证据为中心的表示。该模型在每一步复用同一共享模块，向更早历史延伸扫描不增加参数，并保留扫描中的中间状态供预测头分别权衡短期与长期历史。在多个真实时间序列数据集上，MARO 在长期与短期预测上均达到 SOTA。
正文
Abstract:Long-term forecasting models commonly process all patches in a look-back window using the same fixed stack. Older contextual patches and recent evidence therefore receive the same computational depth. Yet the information closest to the forecast and the more distant context do not contribute equally. Uniform processing leaves this distinction unexpressed in the architecture. We propose MARO, a Most-Recent Anchoring with Recurrent Ordering model that processes the look-back window from the most recent patch to the oldest. The most recent patch serves as the anchor. It initializes the latent state and conditions each subsequent step, so older patches are folded into a representation that remains centered on recent evidence. A single shared module is reused at every step, so extending the scan further into the past introduces no additional parameters. Intermediate states retained during the scan allow the forecast head to weigh short and long portions of the history separately. This expresses recency through the order of recurrent refinement. Extensive experiments across multiple real-world time series datasets show that MARO achieves state-of-the-art performance on both long-term and short-term forecasting this http URL studies examine the contribution of the main architectural components.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.03494 [cs.LG] |
| (or arXiv:2610.03494v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.03494 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jung Min Choi [view email]
[v1]
Fri, 2 Oct 2026 15:55:56 UTC (58 KB)
来源：arXiv：cs.LG · arxiv.org