Back to feed
arXiv cs.LG
arXiv cs.LG
7/22/2026
Beyond Output-Space Calibration: Spectral Evidence Bundling for Selective Reliability Estimation in Time-Series Classification

Beyond Output-Space Calibration: Spectral Evidence Bundling for Selective Reliability Estimation in Time-Series Classification

Short summary

This paper introduces a validation-gated reliability policy for time-series classification that combines output confidence with spectral evidence (band energy, entropy, peak dominance, phase stability) to estimate whether a prediction should be trusted. A validation gate enables spectral conditioning only when it improves correctness ranking without breaching safety tolerances. Across eight UCR/UEA datasets and eight backbone families, the method raises Corr-AURC from 0.693 to 0.786 and reduces [email protected] to 0.094.

  • •Proposes spectral evidence bundling for time-series classifier reliability estimation
  • •Validation gate activates spectral conditioning only when it improves safety metrics
  • •Raises Corr-AURC from 0.693 to 0.786 across eight UCR/UEA datasets and eight backbones

Generated with AI, which can make mistakes.

#research-breakthrough#open-source
Read full article at arXiv cs.LG

Is this a good recommendation for you?

Comments

Failed to load comments. Please try again.

Explore more

FALCON-Discover: Discovering Concentrated False-Confidence Regions for Calibration
arXiv cs.LG
Calibrated Selective Fact-Checking via Evidence Chain Evaluation
arXiv CS.AI
Uncertainty-Aware Trust Estimation for Multi-LLM Systems via Structured Expert Judgement
arXiv cs.LG
Uncertainty-Aware Sequential Decision Rules for Event-Triggered LLM Invocation in Streaming Systems
arXiv cs.LG