analiza aiqbench
This paper frames a practical problem in evidence-grounded NLP, where verifying one text span constrains which nested spans are worth checking, as online isotonic regression over a 2D product order. It gives regret bounds for this setting and reports fewer verifier calls on QA and fact-verification benchmarks such as Natural Questions, SciFact, and ERASER. For ML readers, the main points are the theoretical framing and the empirical gains from using containment structure to schedule expensive verification, though the results are reported on a limited set of benchmarks and backbones.
Online Evidence-Containment Prediction over Text Spans under Distinct Queries
arXiv:2610.41384v1 Announce Type: new Abstract: Evidence-grounded NLP systems often verify nested text regions sequentially, so verifying one region constrains others. We formalize this as Online Evidence-Containment Prediction (OECP): each candidate is predicted before exact verification and queried at most once. Span containment is exactly a triangular 2D product order, making OECP distinct-query online isotonic regression. With multiple admissible targets, worst-case regret is near-linear, $\widetilde{\Theta}(m_{\mathrm{eff}})$, in the effective scale given by endpoint-rank compression; a single fixed target, as in standard QA, admits an explicit polynomial-time learner with optimal $\Theta(\log m_{\mathrm{eff}})$ regret, run on real NQ episodes. On Natural Questions, containment yields a large deductive gain plus a smaller learned gain on unresolved candidates, reproduced across two backbones and DPR hierarchies. Expected Search-space Reduction (ESR) turns these probabilities into decisions, localizing evidence in fewer verifier calls on SciFact, ERASER, and DPR hierarchies; on DPR hierarchies under call budgets it outperforms four recent external selectors and raises fixed-reader F1 over learned and generator-greedy scheduling.
Abstract:Evidence-grounded NLP systems often verify nested text regions sequentially, so verifying one region constrains others. We formalize this as Online Evidence-Containment Prediction (OECP): each candidate is predicted before exact verification and queried at most once. Span containment is exactly a triangular 2D product order, making OECP distinct-query online isotonic regression. With multiple admissible targets, worst-case regret is near-linear, $\widetilde{\Theta}(m_{\mathrm{eff}})$, in the effective scale given by endpoint-rank compression; a single fixed target, as in standard QA, admits an explicit polynomial-time learner with optimal $\Theta(\log m_{\mathrm{eff}})$ regret, run on real NQ episodes. On Natural Questions, containment yields a large deductive gain plus a smaller learned gain on unresolved candidates, reproduced across two backbones and DPR hierarchies. Expected Search-space Reduction (ESR) turns these probabilities into decisions, localizing evidence in fewer verifier calls on SciFact, ERASER, and DPR hierarchies; on DPR hierarchies under call budgets it outperforms four recent external selectors and raises fixed-reader F1 over learned and generator-greedy scheduling.
Artificial Intelligence (cs.AI)
arXiv:2610.41384 [cs.AI]
(or arXiv:2610.41384v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.41384 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 250 cuvinte