GITCO: Gated Inference-Time Context Optimization in TSFMs

GITCO：时间序列基础模型（TSFMs）中的门控推理时上下文优化

Abstract: Patch-based Time Series Foundation Models (TSFMs) suffer from context poisoning: structurally anomalous patches capture disproportionate attention and silently degrade zero-shot forecast quality. 摘要： 基于补丁（Patch-based）的时间序列基础模型（TSFMs）常受“上下文中毒”困扰：结构异常的补丁会捕获不成比例的注意力，从而在无声中降低零样本预测的质量。

We propose improving TSFM accuracy at inference time by optimizing the input context rather than modifying model weights. 我们提出在推理阶段通过优化输入上下文来提高 TSFM 的准确性，而不是修改模型权重。

We present GITCO (Gated Inference-Time Context Optimization), a lightweight three-component framework: Gate, Router, and Critic that selectively identifies and suppresses harmful patches without any parameter updates. 我们提出了 GITCO（门控推理时上下文优化），这是一个轻量级的三组件框架，包含门控（Gate）、路由（Router）和评估（Critic），能够在不更新任何参数的情况下，选择性地识别并抑制有害补丁。

Evaluated on TimesFM 2.5 across 53 GIFT-Eval datasets under K-fold cross-validation, GITCO achieves an average +1.95% MASE reduction on TimesFM 2.5 while capturing 89.9% of the improvement upper bound. 在 53 个 GIFT-Eval 数据集上通过 K 折交叉验证对 TimesFM 2.5 进行评估，GITCO 在 TimesFM 2.5 上实现了平均 1.95% 的 MASE（平均绝对比例误差）降低，同时捕获了 89.9% 的改进上限。

We introduce context sensitivity profiles as a new characterizable property of TSFMs: the mapping from time series meta-features to expected accuracy improvement under inference-time context intervention, shaped jointly by model architecture and the statistical structure of the data. 我们引入了“上下文敏感性概况”（context sensitivity profiles）作为 TSFM 的一种可表征新属性：即从时间序列元特征到推理时上下文干预下预期准确性提升的映射，该映射由模型架构和数据的统计结构共同塑造。