Learnijoy
ResearchAI ResearchAI Engineering & DevTools

SMart Boosts Time Series Learning with Multi-Source Transfer.

Fang He, Wang-chien Lee· September 3, 2026 View original

Key takeaways

  • SMart improves time series representation learning through multi-source and multi-phase pre-training.
  • The framework uses recurrence plots to capture time series dynamics effectively.
  • A source dataset selector enhances knowledge transfer from diverse datasets.
  • It significantly outperforms state-of-the-art models in classification and regression tasks.

Who benefits

FinanceHealthcareManufacturingIoTRetail

Summary

This paper introduces SMart, a new framework for time series representation learning that uses a multi-phase recurrence plots recovery task and a source dataset selector. It leverages multiple external datasets to enhance pre-training and significantly improves performance in time series classification and regression.

Time series analysis is a critical area in machine learning, and recent advancements have explored using transformer architectures and transferring knowledge from other datasets. This new framework, called SMart, builds upon these ideas by introducing two novel mechanisms to improve time series representation learning. First, SMart employs a multi-phase recurrence plots recovery task, which guides the encoder to embed dynamic patterns into the time series representation. Second, it includes a source dataset selector that intelligently picks multiple suitable external datasets to supplement the target dataset during the pre-training phase. This multi-source, multi-phase approach allows SMart to outperform existing state-of-the-art models, showing significant improvements in accuracy for classification and reduced error for regression across various time series datasets.

Why it matters

Data scientists and AI engineers working with time series data can achieve higher accuracy and more robust models by leveraging this framework, especially in scenarios with limited target data.

How to implement this in your domain

  1. 1Evaluate SMart's performance on existing time series datasets within your organization.
  2. 2Explore integrating the multi-phase recurrence plots recovery task into custom time series models.
  3. 3Develop strategies for identifying and curating suitable multi-source datasets for pre-training.
  4. 4Apply the SMart framework to improve predictive models in areas like anomaly detection or forecasting.

Original post by Fang He, Wang-chien Lee

"arXiv:2609.02203v1 Announce Type: new Abstract: Time series representation learning (TSRL) has attracted growing research interests in recent years. Two recent explorations in TSRL are: i) exploiting a transformer-based framework to learn time series; ii) instead of using only th…"

View on X

Originally posted by Fang He, Wang-chien Lee on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

Single AI Model Achieves Robustness Across All Threat Levels

Researchers propose the Threat Conditional Network (TCN), a single AI model that achieves strong adversarial robustness across a continuous range of threat levels. TCN uses a threat-invariant backbone and a lightweight threat-conditional adaptor, matching or surpassing ensembles of specialized models with minimal overhead.

Zhichao Hou, Xiaorui LiuSep 3, 2026
AI ResearchAI Engineering & DevTools

New Broad Learning System Boosts Robustness with Fuzzy Wave Loss

Researchers introduce IFW-BLS, an Intuitionistic Fuzzy Wave Broad Learning System, designed to be robust against both large residuals from noise/outliers and unreliable samples. It achieves this by combining a bounded, asymmetric wave loss with intuitionistic fuzzy scores for sample credibility.

Mushir Akhtar, M. TanveerSep 3, 2026
AI ResearchAI Engineering & DevTools

Multi-Turn AI Agents Need Coverage, Not Just Targeted Credit

This research argues that for multi-turn AI agents, credit assignment should prioritize "coverage" of the causal chain rather than "targeting" specific turns, especially when verifier information density is low. Uniform reward distribution often outperforms sparse, targeted rewards in such scenarios.

Chenyu Zhou, Qiliang Jiang, Shuning Wu, Xu ZhouSep 3, 2026