DECIDER:利用基础模型先验改进模型故障检测与解释
DECIDER: Leveraging Foundation Model Priors for Improved Model Failure Detection and Explanation
AI总结:
本文提出 DECIDER,利用 LLM/VLM 先验构建去偏分类器并依据模型不一致检测图像分类故障,同时通过属性消融给出解释,在多类偏移基准上取得最优检测性能。
摘要:
可靠地检测已部署的机器学习模型在给定输入上何时可能失败,对于确保安全运行至关重要。在本工作中,我们提出 DECIDER(Debiasing Classifiers to Identify Errors Reliably,去偏分类器以可靠识别错误),这是一种利用大型语言模型(LLM)和视觉-语言模型(VLM)的先验来检测图像分类模型故障的新方法。DECIDER 利用 LLM 指定与任务相关的核心属性,并通过使用 VLM 将分类器的视觉特征与这些核心属性对齐,构建分类器的“去偏”版本;随后通过衡量原始模型与去偏模型之间的不一致来检测潜在故障。除了主动识别模型会在其上失败的样本外,DECIDER 还通过一种新颖的属性消融策略,为故障提供人类可解释的解释。通过在涵盖子群体偏移(虚假关联、类别不平衡)和协变量偏移(合成损坏、域偏移)的多种基准上进行大量实验,DECIDER 持续取得最先进的故障检测性能,在总体 Matthews 相关系数以及故障召回率和成功召回率方面显著优于基线。我们的代码可在 https://github.com/kowshikthopalli/DECIDER/ 获取。
英文摘要:
Reliably detecting when a deployed machine learning model is likely to fail on a given input is crucial for ensuring safe operation. In this work, we propose DECIDER (Debiasing Classifiers to Identify Errors Reliably), a novel approach that leverages priors from large language models (LLMs) and vision-language models (VLMs) to detect failures in image classification models. DECIDER utilizes LLMs to specify task-relevant core attributes and constructs a ``debiased'' version of the classifier by aligning its visual features to these core attributes using a VLM, and detects potential failure by measuring disagreement between the original and debiased models. In addition to proactively identifying samples on which the model would fail, DECIDER also provides human-interpretable explanations for failure through a novel attribute-ablation strategy. Through extensive experiments across diverse benchmarks spanning subpopulation shifts (spurious correlations, class imbalance) and covariate shifts (synthetic corruptions, domain shifts), DECIDER consistently achieves state-of-the-art failure detection performance, significantly outperforming baselines in terms of the overall Matthews correlation coefficient as well as failure and success recall. Our codes can be accessed at~\url{https://github.com/kowshikthopalli/DECIDER/}
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。