Research map / Time series / Foundation models for time series
Mantis: Lightweight Foundation Model for Time Series Classification
Vasilii Feofanov, Songkang Wen, Shifeng Xie, Simon Roschmann, Marius Alonso, Hongbo Guo, Romain Ilbert, Malik Tiomoko, Quentin Bouniot, Zeynep Akata, Lujia Pan, Jianfeng Zhang and 1 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| concatenate CLS token with mean of other tokens (intermediate layer) (ablation) | Mantis · UCR | accuracy: +0.8% | Improved |
| same token concatenation at the last layer (ablation) | Mantis · UCR | accuracy | No clear effect |
| self-ensembling over multiple interpolation lengths (ablation) | Mantis · UCR | accuracy: +0.55% | Improved |
| add first-order difference embeddings to self-ensembling (ablation) | SE-Mantis · UCR | accuracy: +0.29% | Improved |
| self-ensembling on the 91-dataset UCR subset (ablation) | SE-Mantis · UCR-91 | accuracy | No clear effect |
| use intermediate layer (layer 3) instead of final layer (ablation) | Mantis · UCR (1M synthetic samples) | accuracy: 0.78 -> 0.7994 | Improved |
| normalize features (Standard scaler) before logistic regression probe (ablation) | Mantis · UCR | accuracy: 0.763 -> 0.837 | Improved |
| replace contrastive pre-training with masked autoencoder (ablation) | Mantis · UCR | accuracy: 0.788 -> 0.716 | Hurt |
| replace InfoNCE with triplet loss (ablation) | Mantis · UCR | accuracy: 0.788 -> 0.767 | Hurt |
| pre-train on synthetic CauKer data instead of equal-size real data (ablation) | Mantis · UCR | accuracy: 0.7829 -> 0.7881 | Improved |
| use modern transformer (RoPE, RMSNorm, SwiGLU) instead of classical (ablation) | Mantis · UCR | accuracy: 0.7989 -> 0.7994 | Improved |
| add first-order difference branch to token generator (ablation) | Mantis · UCR | accuracy | Improved |
Later papers that compared with this method
1 later accepted paper reports Mantis as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- A Unified Shape-Aware Foundation Model for Time Series Classification (AAAI 2026 · better on 2 of 2 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TiReX — TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning (NeurIPS 2025) · better on 2 of 2 shared results
- Catch22+ — CATCH: Channel-Aware Multivariate Time Series Anomaly Detection via Frequency Patching (ICLR 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Enhancing Foundation Models for Time Series Forecasting via Wavelet-based Tokenization (ICML 2025 · similarity 0.87)
- Exploring Representations and Interventions in Time Series Foundation Models (ICML 2025 · similarity 0.85)
- MOMENT: A Family of Open Time-series Foundation Models (ICML 2024 · similarity 0.85)
- A Closer Look at Transformers for Time Series Forecasting: Understanding Why They Work and Where They Struggle (ICML 2025 · similarity 0.85)
- TEMPO: Prompt-based Generative Pre-trained Transformer for Time Series Forecasting (ICLR 2024 · similarity 0.85)
- iTransformer: Inverted Transformers Are Effective for Time Series Forecasting (ICLR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
