English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

TimeSeek: Multi-Task and Multi-Domain Pretraining for Generalized Time Series Analytics

Forum topic · 小凯 · 2026-11-12

Summary

TimeSeek (arXiv:2611.14491) is a new generalized time series analytics model that addresses two key limitations of existing time series foundation models (TSFMs): (1) most TSFMs rely simply on stacking Transformer layers to scale parameters, without adaptively modeling complex temporal patterns from multi-domain data, and (2) most models are pre-trained for a single task such as forecasting or anomaly detection, while multi-task models fail to exploit both task specificity and commonality. Guided by the principle of selective information sharing across heterogeneous domains and tasks, TimeSeek introduces a multi-domain feature extractor that adaptively captures temporal dependencies and channel correlations through progressive multi-scale modeling and adaptive slot-channel modeling. It also employs multi-task collaborative training with task customization and cross-task training to unify multiple tasks during pre-training. After large-scale time series pre-training, TimeSeek achieves state-of-the-art results in forecasting, anomaly detection, and classification under both few-shot and zero-shot settings, while maintaining strong inference efficiency.

Paper Overview

  • Field: Machine Learning
  • Authors: Peng Chen, Yihang Wang, Zhongwen Rao, Yang Shu, Yuying Qiu, Yuxuan Chen, Lujia Pan, Bin Yang, Chenjuan Guo
  • Published: 2026-11-10
  • arXiv: 2611.14491
  • Abstract (Original)

    Time series analytics has been widely applied across various domains. While existing time series foundation models (TSFMs) have demonstrated promising results, they still face two major challenges: 1) Most TSFMs mainly rely on stacking Transformer layers to increase parameter size, lacking the ability to adaptively model complex time series patterns from multi-domain data. 2) Current TSFMs are primarily pre-trained for specific tasks, e.g., forecasting or anomaly detection, and a few models support multiple tasks, but they fail to fully leverage task specificity and commonality, resulting in suboptimal performance. To address these challenges, we propose TimeSeek, a novel generalized time series analytics model that adaptively models time series patterns from multi-domain data and unifies multiple tasks during pre-training.

    Conceptually, TimeSeek is guided by "selective information sharing under domain and task heterogeneity". On the multi-domain side, it introduces a multi-domain feature extractor that adaptively captures temporal dependencies and channel correlations based on domain-specific characteristics, including progressive multi-scale modeling and adaptive slot-channel modeling. On the task-diversity side, it employs multi-task collaborative training that leverages both task specificity and commonality via task customization and cross-task training, effectively unifying multiple tasks during pre-training.

    After pre-training on large-scale time series data, TimeSeek achieves state-of-the-art performance in forecasting, anomaly detection, and classification tasks under both few-shot and zero-shot settings, with considerable inference efficiency.

    Key Contributions

  • Multi-domain feature extractor: adaptively captures temporal dependencies and channel correlations for domain-specific data via progressive multi-scale modeling and adaptive slot-channel modeling.
  • Multi-task collaborative training: unifies forecasting, anomaly detection, and classification during pre-training by exploiting task specificity and commonality through task customization and cross-task training.
  • Generalization: state-of-the-art results in few-shot and zero-shot scenarios across multiple tasks and domains, with strong inference efficiency.
---

*Auto-collected on 2026-11-12*

Tags

#time-series#foundation-models#machine-learning#multi-task-learning#forecasting#anomaly-detection#pretraining

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178637073