跳转至

AsyTO:用于参数高效的多变量时间序列预测的不对称时间算子

文章背景与核心概要

多变量时间序列预测长期面临参数效率与模型灵活性之间的困境:在所有变量间共享同一个时间预测器可以提高参数效率,但迫使异构变量遵循完全相同的历史到未来映射;而为每个变量学习独立的预测器虽能恢复灵活性,但其计算和存储成本会随着变量数量、上下文长度和预测视界的乘积剧烈增长。

本文作者提出了一种名为 AsyTO(Asymmetric Temporal Operator,不对称时间算子) 的创新方法。该方法巧妙地将优化焦点从“压缩观测序列”转变为“压缩预测算子本身”。通过采用不对称、有方向性的历史到未来映射,并对每个变量的算子张量进行因式分解,AsyTO 实现了计算量和参数量随变量数量呈线性扩展。

在包含 11 个基准数据集的大量实验中,AsyTO 展现出卓越的性能,在多数设置下稳定达到了准确率与计算效率的帕累托前沿(Pareto frontier),为轻量化多变量时间序列预测树立了新的技术标杆。


Summary

Multivariate time-series forecasting often struggles with a trade-off between parameter efficiency (sharing predictors across variables) and model flexibility (independent predictors per variable). The authors introduce AsyTO (Asymmetric Temporal Operator), a novel approach that shifts the focus from compressing the observed series to compressing the forecasting operator itself.

By utilizing an asymmetric, directional history-to-future mapping and factorizing the tensor of per-variable operators, AsyTO achieves linear scaling in compute and parameters relative to the number of variables. The model demonstrates superior performance across eleven benchmarks, consistently reaching the accuracy-compute Pareto frontier.

多变量时间序列预测往往在参数效率(在变量间共享预测器)和模型 flexibility(每个变量使用独立的预测器)之间挣扎。作者引入了 AsyTO(不对称时间算子,Asymmetric Temporal Operator),这是一种新颖的方法,将关注点从压缩观测序列转移到压缩预测算子本身。通过利用不对称、有方向性的历史到未来映射,并对每个变量的算子张量进行因式分解,AsyTO 实现了计算量和参数量相对于变量数量的线性扩展。该模型在 11 个基准测试中表现出卓越的性能, consistently 达到了准确率-计算量的帕累托前沿。


Abstract

Multivariate time-series forecasting faces a structural dilemma: sharing one temporal predictor across variables is parameter-efficient but forces heterogeneous variables through an identical history-to-future map, whereas learning an independent predictor per variable restores flexibility at a cost that grows with the product of variable count, context length, and horizon.

We argue that this dilemma dissolves once the object being compressed is the forecasting operator rather than the observed series. Auditing per-variable linear history-to-future maps across standard benchmarks, we find that a phase-locked seasonal component paired with a compact residual operator outperforms a dense phase-blind reference in most audited settings. The residual transport is also directional: lag-invariant alternatives consistently underperform asymmetric history-to-future maps.

Guided by this structure, we propose AsyTO, an Asymmetric Temporal Operator that factorizes the tensor of per-variable operators into shared but distinct history-reading and future-writing temporal modes with per-variable mode-wise gains, complemented by a low-rank periodic prototype and a cycle-separable factorization of the temporal modes. Each forecast reads only its own variable's history, so parameters and compute grow linearly in the number of variables. Across eleven benchmarks and multiple forecast horizons, AsyTO attains the best lightweight error in 30 of 44 dataset-horizon settings, locating at the accuracy-compute Pareto frontier.

多变量时间序列预测面临着一个结构性困境:在各个变量之间共享一个时间预测器具有较高的参数效率,但它迫使异构变量通过完全相同的历史到未来映射;而为每个变量学习一个独立的预测器虽然恢复了灵活性,但其代价随变量数量、上下文长度和预测视界的乘积而增长。

我们认为,一旦被压缩的对象是预测算子而不是观测序列,这个困境就会迎刃而解。通过对标准基准测试中的逐变量线性历史到未来映射进行审查,我们发现,在大多数审查设置中,相位锁定的季节性组件与紧凑的残差算子相结合,其表现优于密集的相位盲参考。残差传输也是有方向性的:滞后不变的替代方案在性能上始终不如不对称的历史到未来映射。

在这种结构的指导下,我们提出了 AsyTO(不对称时间算子),它将逐变量算子的张量因式分解为共享但独立的历史读取和未来写入时间模式,并辅以逐变量的模式增益、低秩周期原型以及时间模式的周期可分因式分解。每个预测只读取其自身变量的历史记录,因此参数和计算量随变量数量呈线性增长。在 11 个基准测试和多个预测视界中,AsyTO 在 44 种数据集-视界设置中的 30 种里获得了最佳的轻量级误差,位于准确率-计算量的帕累托前沿上。


Access & Resources

访问与资源


Metadata

  • Comments: 8 pages, 4 figures, 4 tables
  • Submission History: [v1] Mon, 17 Aug 2026 04:42:16 UTC

元数据

  • 评论: 8 页,4 张图表,4 个表格
  • 提交历史: [v1] 2026年8月17日 星期一 04:42:16 UTC