arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2510.21998cs.LGcs.AIstat.ML
收藏

从黑盒到因果盒:构建更具可解释性的模型

From Black-box to Causal-box: Towards Building More Interpretable Models

Causal Artificial Intelligence Lab(因果人工智能实验室);Columbia University(哥伦比亚大学)

Inwoo Hwang, Yushu Pan, Elias Bareinboim

AI总结:

本文提出因果可解释性概念,分析黑盒与概念模型均不满足,并开发框架设计因果可解释模型,通过图准则权衡可解释性与准确性,实验验证理论。

摘要:

理解深度学习模型做出的预测仍然是一个核心挑战,尤其是在高风险应用中。一个有前景的方法是赋予模型回答反事实问题的能力——这些假设性的“如果……会怎样?”场景超越了观测数据,能够洞察模型的推理过程。在这项工作中,我们引入了因果可解释性的概念,它形式化了何时可以从特定类别的模型和观测数据中评估反事实查询。我们分析了两种常见的模型类别——黑盒预测器和基于概念的预测器——并表明这两种模型在一般情况下都不具有因果可解释性。为了解决这一差距,我们开发了一个框架,用于构建在设计上就具有因果可解释性的模型。具体来说,我们推导出一个完整的图准则,用于确定给定的模型架构是否支持给定的反事实查询。这导致了因果可解释性与预测准确性之间的基本权衡,我们通过识别唯一的、能产生具有最大预测表达力的可解释模型的最大特征集来刻画这一权衡。实验证实了理论发现。

英文摘要:

Understanding the predictions made by deep learning models remains a central challenge, especially in high-stakes applications. A promising approach is to equip models with the ability to answer counterfactual questions -- hypothetical ``what if?'' scenarios that go beyond the observed data and provide insight into a model reasoning. In this work, we introduce the notion of causal interpretability, which formalizes when counterfactual queries can be evaluated from a specific class of models and observational data. We analyze two common model classes -- blackbox and concept-based predictors -- and show that neither is causally interpretable in general. To address this gap, we develop a framework for building models that are causally interpretable by design. Specifically, we derive a complete graphical criterion that determines whether a given model architecture supports a given counterfactual query. This leads to a fundamental tradeoff between causal interpretability and predictive accuracy, which we characterize by identifying the unique maximal set of features that yields an interpretable model with maximal predictive expressiveness. Experiments corroborate the theoretical findings.

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。

↑