aiqbench analysis
This research introduces LOTUS, a novel pretraining principle called "forecast grounding" that forces time-series language models to retain predictive temporal information before adapting to downstream tasks. By combining raw patches with spectral representations, the method achieves state-of-the-art performance on multiple datasets while using significantly fewer trainable parameters than existing baselines. The approach effectively bridges the gap between linguistic alignment and the preservation of underlying signal evolution in time-series data.
Lotus: Learning to Understand Time-series by Learning to Forecast
arXiv:2610.18877v1 Announce Type: cross Abstract: Time-series language models typically align numerical signals with language using labels, captions, or question-answer pairs. Although such supervision encourages linguistic representations, it does not explicitly require them to preserve information about how the underlying time-series signal evolves. We introduce forecast grounding, a pretraining principle that requires language-aligned time-series representations to remain predictive of the future evolution before they are adapted to downstream tasks. We instantiate this principle in LOTUS, which divides each time-series channel into non-overlapping patches and maps them into the LLM embedding space. It then predicts the next patch from the LLM hidden states of the time-series tokens, encouraging the learned representation to retain temporal information. To capture both temporal changes and periodic patterns, we also propose a temporal-spectral patch representation that combines each raw time-series patch with its spectral representation before mapping it into the LLM embedding space. Across three language model backbones and six downstream evaluation datasets, LOTUS achieves the best average performance with minimal trainable parameters, reaching 59.91% accuracy and outperforming the strongest matched baseline by 3.36%. Our representation analyses further show that LOTUS uses the numerical input and preserves temporal structure in its learned patch representations. These results establish forecast grounding as a promising pretraining principle, enabling compact time-series language models to outperform models 25 to 50 times their size on TSRBench.
Abstract:Time-series language models typically align numerical signals with language using labels, captions, or question-answer pairs. Although such supervision encourages linguistic representations, it does not explicitly require them to preserve information about how the underlying time-series signal evolves. We introduce forecast grounding, a pretraining principle that requires language-aligned time-series representations to remain predictive of the future evolution before they are adapted to downstream tasks. We instantiate this principle in LOTUS, which divides each time-series channel into non-overlapping patches and maps them into the LLM embedding space. It then predicts the next patch from the LLM hidden states of the time-series tokens, encouraging the learned representation to retain temporal information. To capture both temporal changes and periodic patterns, we also propose a temporal-spectral patch representation that combines each raw time-series patch with its spectral representation before mapping it into the LLM embedding space. Across three language model backbones and six downstream evaluation datasets, LOTUS achieves the best average performance with minimal trainable parameters, reaching 59.91% accuracy and outperforming the strongest matched baseline by 3.36%. Our representation analyses further show that LOTUS uses the numerical input and preserves temporal structure in its learned patch representations. These results establish forecast grounding as a promising pretraining principle, enabling compact time-series language models to outperform models 25 to 50 times their size on TSRBench.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2610.18877 [cs.LG]
(or arXiv:2610.18877v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.18877 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 317 words
Related coverage
News
Quantum Computing Report · Jul 25, 2026
7 Steps to Mastering Time Series Analysis with Python
KDnuggets · Jun 4, 2026
Presentation: Designing AI Platforms for Reliability: Tools for Certainty, Agents for Discovery
InfoQ · May 27, 2026
Prophet vs NeuralProphet vs TimeGPT vs Chronos: A Practical Comparison
Analytics Vidhya · Jun 9, 2026