SABRE: A Multi-Agent Approach for Selecting Out-of-Distribution Detectors Under a Budget
3.00T1 sourcearXiv cs.MA
Source record
Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2608.02959.
Pipeline notes
The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.
SummaryPaper introduces SABRE, a three-agent system (Selector, Reporter, Analyst) that dynamically chooses out-of-distribution detectors for vision-language models at inference time under a query budget, showing that fixed detector choices invert across domains and per-regime selection recovers reliable detection.
Why it mattersDocuments a concrete failure mode (detector inversion across domains) and a budgeted three-agent selection pattern worth noting for production VLMs and agent orchestration design.
Cited by
No citations on record.