At the same time that AI and machine learning are becoming central to human life, their potential harms become more vivid. In the presence of such drawbacks, a critical question one needs to address before using these data-driven technologies to make a decision is whether to trust their outcomes. Aligned with recent efforts on data-centric AI, this paper proposes a novel approach to address the trust question through the lens of data, by associating data sets with distrust quantification that specify their scope of use for predicting future query points. The distrust values raise warning signals when a prediction based on a dataset is questionable and are valuable alongside other techniques for trustworthy AI. We propose novel algorithms for computing the distrust values in the neighborhood of a query point efficiently and effectively. Learning the necessary components of the measures from the data itself, our sub-linear algorithms scale to very large and multi-dimensional settings. Besides demonstrating the efficiency of our algorithms, our extensive experiments reflect a consistent correlation between distrust values and model performance. This underscores the message that when the distrust value of a query point is high, the prediction outcome should be discarded or at least not considered for critical decisions.
翻译:在AI和机器学习成为人类生活的核心的同时,它们的潜在危害会变得更加生动。在出现这些缺陷的情况下,在使用这些数据驱动的技术来作出决定之前,一个关键问题需要解决,这就是是否信任它们的结果。 与最近关于以数据为中心的AI的努力相一致,本文件提出了一种新颖的方法,通过数据透镜来解决信任问题,将数据集与确定它们用于预测未来查询点的范围的不信任量化结合起来。不信任值在基于数据集的预测有疑问时会发出警告信号,并且与其他技术一道对可信赖的AI有价值。我们提出了在查询点附近高效和有效地计算不信任值的新算法。从数据本身、我们的次线性算法规模到非常大和多维环境学习衡量的必要组成部分。除了展示我们的算法的效率外,我们的广泛实验还反映了不信任值和模型性能之间的一贯关联。这强调了这样的信息,即当一个查询点的不信任值很高时,预测结果应该被抛弃,或者至少不考虑批评性决定。