Skip to content

Scaling and Generalizing Time-Series Cleaning

Home / Open Questions / Scaling and Generalizing Time-Series Cleaning

Background: Multivariate time series (MTS) cleaning involves addressing diverse and co-occurring quality issues like missing values, outliers, and constraint violations without relying on ground truth data. Existing systems for automated pipeline generation often struggle with generalization to unseen domains and the computational burden of searching large combinatorial operator spaces.

Question / Future Work: Further research is required to extend hierarchical agent-based cleaning systems to real-time streaming data environments, where latency and data volume present significant challenges, and to investigate integrating time series foundation models to enhance adaptability across domains.

Why It Matters: Addressing these challenges would transition automated cleaning from offline batch processing to real-time operational environments and significantly reduce the need for dataset-specific tuning.

Metadata & Links