Rethinking Zero-Shot Time Series Classification: From Task-specific Classifiers to In-Context Inference
For researchers and practitioners in time series analysis, this work provides a training-free, in-context inference method that eliminates classifier-dependent bias in zero-shot evaluation.
The paper tackles the issue of evaluation bias in zero-shot time series classification caused by task-specific classifiers. It proposes TIC-FM, an in-context learning framework that achieves strong accuracy on 128 UCR datasets, with consistent gains in extreme low-label settings, without parameter updates.
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 classifier-dependent 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 encoder and a lightweight projection adapter with a split-masked 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.