Rethinking Zero-Shot Time Series Classification: From Task-specific Classifiers to In-Context Inference
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866917238618980352 |
|---|---|
| author | Fang, Juntao Xie, Shifeng Nie, Shengbin Ling, Yuhui Liu, Yuming Li, Zijian Zhang, Keli Pan, Lujia Palpanas, Themis Cai, Ruichu |
| author_facet | Fang, Juntao Xie, Shifeng Nie, Shengbin Ling, Yuhui Liu, Yuming Li, Zijian Zhang, Keli Pan, Lujia Palpanas, Themis Cai, Ruichu |
| contents | 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 |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_00620 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Rethinking Zero-Shot Time Series Classification: From Task-specific Classifiers to In-Context Inference Fang, Juntao Xie, Shifeng Nie, Shengbin Ling, Yuhui Liu, Yuming Li, Zijian Zhang, Keli Pan, Lujia Palpanas, Themis Cai, Ruichu Machine Learning Artificial Intelligence 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 |
| title | Rethinking Zero-Shot Time Series Classification: From Task-specific Classifiers to In-Context Inference |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2602.00620 |