arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2501.14136cs.LG
收藏

显著性图具有歧义性:一阶与二阶归因的逻辑关系分析

Saliency Maps are Ambiguous: Analysis of Logical Relations on First and Second Order Attributions

Osnabrück University(奥斯纳布吕克大学);German Research Center for Artificial Intelligence (DFKI)(德国人工智能研究中心)

Leonid Schwenke, Martin Atzmueller

AI总结:

本文扩展了基于ANDOR框架的显著性方法分析,通过更多数据集和全局一致性表示评估,揭示其在不同场景下的失效模式,并提出真正的输入省略方法。

摘要:

近期研究揭示了基于归因或热图的显著性方法中存在的潜在缺陷。一个典型缺陷是确认偏差,即分数与人类预期进行比较。由于缺乏真实模型推理的基准真值,衡量显著性方法的质量较为困难,因此发现其普遍局限性也同样困难。此外,基于掩蔽的评估在复杂数据上容易引入偏差,因为大多数方法无法完全忽略输入。在本工作中,我们扩展了先前基于逻辑数据集框架ANDOR的分析,该分析表明所有被分析的显著性方法在所有可能场景下均无法获取所需的全部分类信息。具体而言,本文通过分析更多数据集来扩展先前工作,以更好地理解显著性方法失效的场景。此外,我们应用全局一致性表示作为额外的评估方法,以实现真正的输入省略。

英文摘要:

Recent work uncovered potential flaws in \eg attribution or heatmap based saliency methods. A typical flaw is a confirmations bias, where the scores are compared to human expectation. Since measuring the quality of saliency methods is hard due to missing ground truth model reasoning, finding general limitations is also hard. This is further complicated, because masking-based evaluation on complex data can easily introduce a bias, as most methods cannot fully ignore inputs. In this work, we extend our previous analysis on the logical dataset framework ANDOR, where we showed that all analysed saliency methods fail to grasp all needed classification information for all possible scenarios. Specifically, this paper extends our previous work using analysis on more datasets, in order to better understand in which scenarios the saliency methods fail. Further, we apply the Global Coherence Representation as an additional evaluation method in order to enable actual input omission.

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。

↑