SOURCE-LINKED INTELLIGENCE

Diagnosing Temporal Misalignment in Multichannel Time-Series Classification with Minimum Description Length

arXiv · AI, language, vision and robotics · article · Sep 13, 2026 · UTC

Multichannel time-series classification commonly assumes synchronized sensor streams, although latency, clock drift, and preprocessing can introduce relative delays during data collection or after deployment. Existing synchronization solutions are often hardware-specific and difficult to apply retrospectively. Consequently, synchronization problems may remain undetected while classification performance is suboptimal. We introduce a classifier- and label-free diagnostic based on minimum description length (MDL). Our method applies candidate temporal shifts to sensor groups and measures how effi

Read original source ↗ Open in workspace

recordType

paper

region

Global