Single-Query Black-Box Calibration Auditing via Logit Bias

SummaryAI generated

On September 7, 2026, arXiv cs.LG published the paper “Single-Query Black-Box Calibration Auditing via Logit Bias”, which proposes a new method for calibrating and auditing black-box machine learning models with just a single query. This study evaluates the consistency of the model’s prediction distribution with the true label distribution by analyzing logit bias, aiming to address the issue of low audit efficiency caused by traditional multi-query methods or white-box dependencies. This method significantly reduces computational costs while maintaining high accuracy, providing a new approach for industry to quickly verify the fairness and reliability of models.

Related eventsRELATED EVENTS

2026-09-09 05:29OpenAI claims huge maths breakthrough on a famed ‘Millennium Problem’
2026-09-09 06:11OpenAI claims to have solved the Navier-Stokes equation problem, but its originality is questioned by the academic community
2026-09-08 22:00Google DeepMind Maps 9 Billion Possible DNA Variants

All reports (1)SOURCES

↗

A arXiv cs.LG en 2026-09-07 12:00

研究人员提出了一种通过 logits 偏差参数对大语言模型进行单查询黑盒校准审计的新方法。该方法利用任意暴露 logits_bias 参数的 LLM API，仅需每个样本一次查询即可数学操纵以评估精确概率阈值。基于此机制，研究团队引入了一个针对二元任务的新颖且可证明一致的 True Calibration Error 估计器，从而为审计黑盒基础模型提供了高效框架。