Rethinking Zero-Shot Time Series Classification From Task-specific Classifiers to In-Context Inference

Jul 16, 2026

Upload by Finn

Zero-shot evaluation of time series foundation models (TSFMs) for classification often freezes the encoder and trains or attaches a task-specific classifier, which breaks the training-free assumption and can bias results through classifier-dependent choices. TIC-FM introduces a training-free in-context learning framework that treats the labeled support set as context and predicts test labels in a single forward pass. A split-masked latent-memory Transformer and lightweight adapter enable theory-backed classifier emulation and strong gains on 128 UCR datasets, especially under extreme low-label settings.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages24

Reading Time1 h 0 min

Abstract

Introduction

Related Work

Rethinking Zero-Shot Time Series Classification: From Task-specific Classifiers to In-Context Inference Juntao Fang * 1 Shifeng Xie * 2 3 Shengbin Nie 1 Yuhui Ling 1 Yuming Liu 1 Zijian Li 4 Keli Zhang 2 Lujia Pan 5 Themis Palpanas 3 Ruichu Cai 1 arXiv :2602 .00620v2 [ cs .LG] 13 Jul 2026 Abstract The zero-shot evaluation of time series foundation models (TSFMs) for classification typically uses a frozen encoder followed by a task-specific classifier. However, this practice violates the training-free premise of zero-shot deployment and introduces evaluation bias due to classifierdependent training choices. To address this issue, we propose TIC-FM, an in-context learning framework that treats the labeled training set as context and predicts labels for all test instances in a single forward pass, without parameter updates. TIC-FM pairs a time series encoderand a lightweight projection adapter with a splitmasked latent memory Transformer. We further provide theoretical justification that in-context inference can subsume trained classifiers and can emulate gradient-based classifier training within a single forward pass. Experiments on 128 UCR datasets show strong accuracy, with consistent gains in the extreme low-label situation, highlighting training-free transfer for time series classification.The source code is publicly available at [https://github.com/fangjuntao/TIC-FM](https://github.com/fangjuntao/TIC-FM). 1. Introduction Time series classification is a core task across applications such as human movement analysis, clinical monitoring and diagnosis, digital health, and energy systems, where accurate recognition of temporal patterns directly supports decision-making (Ismail Fawaz et al., 2019) . Recent progress on time series foundation models (TSFMs) makes it increasingly feasible to deploy a single pretrained *Equal contribution 1 Guangdong University of Technology, Guangzhou, China 2Huawei Noah’s Ark Lab, Paris, France 3Paris Descartes University, Paris, France 4Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, United Arab Emirates 5Huawei Noah’s Ark Lab, Shenzhen, China. Correspondence to: Ruichu Cai <[cairuichu@gmail.com](cairuichu@gmail.com) >. Preprint. July 14, 2026. backbone across domains (Liang et al., 2024) . At the sametime, in privacy-sensitive or labor-intensive settings, like healthcare, collecting and annotating labeled time series at scale is often costly or infeasible, which amplifies the need for transfer without extensive task-specific training (Gao et al., 2025) . Consequently, strong zero-shot and few-shot generalization is not merely desirable but central to realizing practical for classification with TSFMs. Despite these aspirations, the evaluation practice for TSFMs remains largely limited to a frozen encoder and task-specific classifier practice. In this setup, the backbone is kept frozen, and a task-specific classifier (e.g., an SVM or an MLP) is trained on top of the extracted embeddings to adapt to downstream datasets. While widely adopted, we argue that this protocol suffers from fundamental methodological shortcomings. First, training task-specific classifiers requires explicit parameter optimization and access to a training set for hyperparameter tuning. This practice conflicts with the zero-shot premise of foundation models, since the model still requires a supervised training step before deployment. Second, the protocol induces substantial evaluation bias and sensitivity. As our empirical analysis shows (see Table 1), the effectiveness of a frozen backbone depends heavily on the choice of task-specific classifier (e.g., SVM vs. MLP) . Such dependence inevitably conflates the intrinsic quality of the learned representations with the expressivity of the classifier, thereby obstructing a fair pipeline of the foundation model itself. Finally, training a task-specific classifier is less reliable in the extremely low-shot regime. Optimizing a high-dimensiona

FAQs

为什么传统的零样本时间序列分类评估需要改进？

传统做法通常冻结编码器并依赖任务特定分类器训练或选择，这与“训练免费”的零样本部署前提冲突，并且分类器选择会引入评估偏差与敏感性，尤其在极低标注场景更不稳定。

TIC-FM如何实现训练免费的一次前向推理？

TIC-FM将带标签的支持集当作上下文提示，不进行任何参数更新；通过同时关注上下文与查询样本，在单次前向传播中推断测试实例的标签。

TIC-FM的关键架构组件是什么？

方法由时间序列编码器、轻量级投影适配器以及带分裂遮罩与潜在记忆的上下文推理Transformer分类器构成；标签只注入支持token以实现泄漏规避的并行推断。

Jul 16, 2026

Upload by Finn