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 2603.22235cs.HCcs.LG
收藏

ShapDBM:在Shapley空间中探索决策边界图

ShapDBM: Exploring Decision Boundary Maps in Shapley Space

School of Computing, Newcastle University, UK(新castle大学计算机学院);Department of Information and Computing Science, Utrecht University, Netherlands(乌得勒支大学信息与计算科学系)

Luke Watkin, Daniel Archambault, Alex Telea

AI总结:

提出通过将数据空间转换为Shapley空间并计算降维来生成决策边界图,相比直接基于数据的方法,生成的图质量指标相似或更高,决策区域更紧凑、更易探索且与模型性能更一致。

摘要:

决策边界图(DBM)是可视化机器学习分类边界的有效工具。然而,DBM的质量很大程度上取决于降维(DR)技术和用于数据点的高维空间。对于复杂的机器学习数据,降维可能会产生许多混合类别,导致DBM难以使用甚至产生误导。我们提出了一种新技术,通过将数据空间转换为Shapley空间并对其计算降维来生成DBM。与直接从数据计算的DBM相比,我们的图具有相似或更高质量指标值,并且决策区域明显更紧凑、更易于探索,与测量的模型性能更一致。

英文摘要:

Decision Boundary Maps (DBMs) are an effective tool for visualising machine learning classification boundaries. Yet, DBM quality strongly depends on the dimensionality reduction (DR) technique and high dimensional space used for the data points. For complex ML data, DR can create many mixed classes which yield DBMs that are hard to use or even misleading. We propose a new technique to compute DBMs by transforming data space into Shapley space and computing DR on it. Compared to DBMs computed directly from data, our maps have similar or higher quality metric values and visibly more compact, easier to explore, decision zones that better agree with measured model performance.

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。

↑