MantisV2: Closing the Zero-Shot Gap in Time Series Classification with Synthetic Data and Test-Time Strategies
Vasilii Feofanov, Songkang Wen, Jianfeng Zhang, Lujia Pan, Ievgen Redko · 2026
Read MantisV2: Closing the Zero-Shot Gap in Time Series Classification with Synthetic Data and Test-Time Strategies by Vasilii Feofanov, Songkang Wen, Jianfeng Zhang, Lujia Pan, Ievgen Redko (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-17868.txt.
Description
Vasilii Feofanov and team introduce MantisV2, closing the zero-shot gap in time series sorting with fake data and test-time tricks. Base models for time series sorting can serve as universal feature tools for varied later tasks, yet a large gap remained between frozen and fine-tuned nets in early models such as Mantis. The paper strengthens zero-shot feature extraction through three moves.
First, Mantis+, a version of Mantis pre-trained fully on fake time series, is introduced. Second, careful cut-out studies refine the design into MantisV2, an improved and light encoder. Third, an enhanced test-time method leverages mid-layer codes and refines output-token pooling, with self-mixing and cross-model vector fusion adding further gains. Tests on UCR, UEA, human activity recognition and EEG sets show MantisV2 and Mantis+ always beating prior time series base models, reaching state-of-the-art zero-shot scores.
Catalog record
| CREATOR | Vasilii Feofanov, Songkang Wen, Jianfeng Zhang, Lujia Pan, Ievgen Redko |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 158 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-17868