arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1909.06342cs.LGcs.AIcs.CYcs.HCstat.ML
收藏

部署中可解释的机器学习

Explainable Machine Learning in Deployment

Carnegie Mellon University(卡内基梅隆大学);Partnership on AI(人工智能合作组织);University of Cambridge(剑桥大学);Leverhulme CFI(Leverhulme智能未来中心);University of Texas at Austin(德克萨斯大学奥斯汀分校);The Alan Turing Institute(艾伦图灵研究所);Fiddler Labs;Baidu(百度);CognitiveScale;IBM Research(IBM研究院)

Umang Bhatt, Alice Xiang, Shubham Sharma, Adrian Weller, Ankur Taly, Yunhan Jia, Joydeep Ghosh, Ruchir Puri, José M. F. Moura, Peter Eckersley

摘要:

可解释的机器学习通过使用各种方法(如特征重要性分数、反事实解释或有影响力的训练数据),为利益相关者提供了对模型行为的洞察。然而,人们对组织如何在实践中使用这些方法知之甚少。本研究探讨了组织如何看待和使用利益相关者消费的可解释性。我们发现,目前,大多数部署不是针对受模型影响的最终用户,而是针对机器学习工程师,他们使用可解释性来调试模型本身。因此,在实践中的可解释性与透明度目标之间存在差距,因为解释主要是为内部利益攸关方服务,而不是为外部利益攸关方服务。我们的研究综合了当前可解释性技术的局限性,这些局限性阻碍了最终用户的使用。为了促进最终用户的互动,我们开发了一个框架,建立明确的目标,可解释性。最后,我们讨论了关于可解释性的问题。

英文摘要:

Explainable machine learning offers the potential to provide stakeholders with insights into model behavior by using various methods such as feature importance scores, counterfactual explanations, or influential training data. Yet there is little understanding of how organizations use these methods in practice. This study explores how organizations view and use explainability for stakeholder consumption. We find that, currently, the majority of deployments are not for end users affected by the model but rather for machine learning engineers, who use explainability to debug the model itself. There is thus a gap between explainability in practice and the goal of transparency, since explanations primarily serve internal stakeholders rather than external ones. Our study synthesizes the limitations of current explainability techniques that hamper their use for end users. To facilitate end user interaction, we develop a framework for establishing clear goals for explainability. We end by discussing concerns raised regarding explainability.

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。

↑