Welcome to AT-MAS
Scientific systems, such as particle accelerators and chemical reactors, continuously generate time series data that reflect their evolving states and are essential for understanding, monitoring, and operating these systems. Turning these data into useful results involves a chain of steps, including data preparation, querying, model selection, analysis, and result checking, which experts still assemble, configure, and revise manually as the data evolve. Agentic AI offers a promising new paradigm for automating this chain, where AI agents plan tasks, use tools, and adjust their actions based on intermediate results. However, building such agents remains challenging: they must reason about temporal dependencies, interpret observations under domain knowledge and physical constraints, coordinate multi-step workflows, and operate reliably at scale.
The goal of this workshop is to bring together researchers and practitioners from time series data management and analytics, machine learning, agentic AI, and scientific domains, to share recent advances, identify key research problems, and connect new methods with real-world applications.
Call for Papers
We invite original research papers on agentic time series data management and analytics for science, including both full and short papers on algorithms, systems, benchmarks, and applications. Topics of interest include, but are not limited to:
- Agentic data preparation for time series, such as cleaning, alignment, imputation, and integration
- Agentic query processing over time series, such as natural-language querying and query optimization
- Agentic workflow planning and tool orchestration for multi-step time series analytics
- Data-centric agentic time series analytics, such as forecasting, anomaly detection, and diagnosis
- Memory management for time series agents, including indexing and retrieval of historical context
- Multi-agent coordination for managing and analyzing distributed and heterogeneous scientific time series data
- Correctness, provenance, verification, and error recovery in agentic workflows
- Efficient and scalable agents for large-scale and streaming scientific time series
- Robustness, explainability, and human oversight in agentic time series data management and analysis
- Physics-informed and domain-aware agents for scientific time series
- Benchmarks and evaluation for agentic time series systems
- Applications of agentic time series systems in scientific domains, such as chemistry and climate science
Important Dates
Submission
- Submission SiteTBA
- Paper Types
- Full papers: up to 12 pages.
- Short papers: up to 6 pages.
- Page limits exclude references and the AI-generated content acknowledgement.
- No appendix is allowed.
- FormatSubmissions must be in PDF format and follow the IEEE conference format required by ICDE 2027.
- OriginalitySubmissions must present original work that has not been published and is not under review at another venue. A submission must also be substantially different from previously published work.
- ReviewThis workshop follows a single-blind review process.
- ReproducibilityAuthors are encouraged to provide supplemental material, such as code and data, to support reproducibility, especially for papers with experimental or benchmark contributions.
- Best Paper AwardA Best Paper Award will be presented at the workshop.
Organizers
Program Committee
- Lu Chen, Zhejiang University, China
- Yeye He, Microsoft Research, USA