arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2606.04074cs.LGcs.AIcs.ITmath.IT
收藏

自适应分块在时间序列预测中比看起来更难

Adaptive Patching Is Harder Than It Looks For Time-Series Forecasting

ICube, University of Strasbourg, Illkirch-Graffenstaden, France(斯特拉斯堡大学ICube研究所,法国伊尔克里奇-格拉夫芬斯坦德);Technical University of Munich(慕尼黑技术大学);FinTech Thrust, The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州)金融科技研究组);Computer Science Department, Hainan Bielefeld University of Applied Sciences(海南比尔费尔德应用科学大学计算机科学系);Cephalgo, Strasbourg, France(法国斯特拉斯堡Cephalgo公司);Heilbronn Data Science Center, Munich Data Science Institute(慕尼黑数据科学研究所海德堡数据科学中心)

Federico Zucchi, Yi Xie, Chao Zhang, Thomas Lampert, Ziyue Li

AI总结:

本文通过理论分析和实验验证,探讨自适应分块在时间序列Transformer中是否优于调优的均匀分块,发现均匀基线在标准基准上具有竞争力,自适应分块的优势有限且依赖于特定方法和数据集。

摘要:

自适应分块是时间序列Transformer最近提出的一个引人注目的方案:在序列局部信息丰富的区域分配更细的分块。本文探究在什么条件下内容自适应分块算子应优于调优的均匀算子。局部异质性本身并不足够:在逐点预测损失下,一个看似复杂的区域并不自动意味着更细的分块会减少损失。我们将分块建模为有预算的比特率分配,并推导出一个显式阈值,动态分块规则必须满足该阈值才能击败调优的均匀基线,然后从局部(二次代理)和全局(模型假设下的强凸界)两方面界定了可实现的改进。由此得出两个结构性结果:在没有耦合约束的情况下,标量局部复杂度无法在常见损失景观下产生非均匀最优;一旦骨干网络训练到其表示感知最优,对齐增益会在调优的均匀分块大小附近崩溃。为了验证这些预测,我们在三种代表性架构上进行了受控隔离研究,用均匀分块大小扫描替换每个自适应机制,同时保持骨干网络、数据和训练协议不变。在标准的长时域预测基准上,验证选择的均匀基线与动态对应物具有竞争力,每个设置的效果集中在零附近,且按数据集汇总后没有一致的方向性优势。我们观察到的较大增益是方法和数据集特定的。因此,自适应分块应针对调优的均匀基线进行评估;其价值取决于是否有一个廉价且可靠的路由信号能够识别出更细的分块实际上在何处减少预测损失。

英文摘要:

Adaptive patching has been proposed for time-series forecasting: allocate finer patches where the sequence looks locally informative. We compare three adaptive patching methods with tuned uniform counterparts that use one fixed patch size throughout the input, while preserving each method's backbone family and shared training protocol. Across the evaluated long-horizon benchmarks, we observe mostly small accuracy differences and no consistent advantage for adaptive patching; uniform variants generally train faster and avoid adaptive routing at inference. Motivated by these observations, we develop a theoretical framework for budgeted patch allocation through a common resolution-density representation of uniform, nonuniform, and overlapping layouts, independently of the patching rule. Within a separable shared-loss model, we derive an exact criterion for when nonuniform allocation improves on uniform allocation at the same budget. We characterize modeled gains locally through a quadratic surrogate and bound them globally under strong convexity and a bounded loss derivative. The analysis distinguishes where resolution is placed from how much it is redistributed: even an allocation that correctly ranks regions by their weights can redistribute too much resolution and increase modeled loss. Without coupling across time, positive regional weights preserve the minimizers of the shared loss. When the shared loss has a common minimizing rate, the remaining absolute headroom for improvement tends to zero as the uniform rate approaches it. These results characterize the roles of regional weighting, placement, and redistribution strength under the stated model. The empirical findings support evaluating adaptive patching against a tuned uniform baseline and weighing any accuracy gain against the additional training and inference cost of adaptive routing.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑