Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

TimeCopilot

Created by
  • Haebom

Author

Azul Garza, Rene e Rosillo

Outline

TimeCopilot is the first open-source agent prediction framework that combines multiple time-series-based models (TSFMs) and large-scale language models (LLMs) through a single, unified API. TimeCopilot automates feature analysis, model selection, cross-validation, and forecast generation, provides natural language explanations, and supports direct questions about the future. It is compatible with both commercial and open-source models and is an LLM-agnostic framework that supports ensembles across diverse forecasting series. Large-scale GIFT-Eval benchmark results demonstrate that TimeCopilot achieves state-of-the-art probabilistic forecasting performance at a low cost. It provides a practical foundation for reproducible, explainable, and accessible agent prediction systems.

Takeaways, Limitations

•
Takeaways:
◦
We present the first open-source agent prediction framework that integrates multiple TSFMs and LLMs.
◦
Improve efficiency through predictive pipeline automation.
◦
Increased usability with natural language explanations and support for direct questions about the future.
◦
Scalability with LLM agnostic and support for various predictive series.
◦
Achieving state-of-the-art performance on the GIFT-Eval benchmark.
◦
Contribute to building reproducible, explainable, and accessible prediction systems.
•
Limitations:
◦
Limitations is not explicitly mentioned in the paper. Additional benchmark testing and performance evaluation on various datasets may be required. Further analysis may be needed to determine the dependency on specific LLMs or TSFMs or to assess potential performance degradation.
👍
Made with Slashpage