首页 > AI前沿 > Where Tree Ensembles Decide: Structural Local Explanation with Decision Predicate Graphs

Where Tree Ensembles Decide: Structural Local Explanation with Decision Predicate Graphs

arXiv机器学习 2026-11-14 00:24 11 阅读 查看原文
用户需要我将一段纯文本的新闻正文,按照特定的规则排版成一个语义化的HTML片段。 规则如下: 1. 只输出HTML片段本身,不要Markdown围栏,不要解释。 2. 每个自然段落用`

`包裹,长文用`

`/`

`划分小节。 3. 列表内容用`
  • `,引用用`
    `,代码用`
    `,重要句子可用``。
    4. 保留原文所有信息,不得增删改内容。
    5. 不要输出``/``/``等文档级标签。
    
    分析正文内容:
    正文是一段关于树集成模型(Tree ensembles)的学术论文摘要或新闻稿。
    内容分为几个逻辑部分:
    - 第一部分:介绍树集成模型在表格数据中的高准确性,但预测计算分散在数百个决策规则中。特征归因和反事实解释存在局限(未指明响应哪些阈值,结构解释丢失了树的信息)。
    - 第二部分:提出局部决策谓词图(local DPG),将样本执行的规则聚集成可验证的计算记录,具有类支持,能重现每个被解释的预测。
    - 第三部分:从规则中选择关键谓词(critical predicate),提供具体的阈值和共享该阈值的树,无需外部解释器。
    - 第四部分:实验结果(16个数据集,4个集成族),比随机和最近阈值谓词更好地分离不同预测的邻居。单次交叉在46-69%的错误预测中恢复真实类,而TreeSHAP排名第一的特征只有29-50%。在植入故障中找到故障阈值的比例为46-71%,随机只有6-21%。
    - 第五部分:尽管这些谓词的免训练编辑没有减少未见数据上的误差,但关键谓词通过指出预测可以改变的位置,补充了归因和反事实。
    
    由于这是一段纯文本,没有明显的列表、引用或代码。我需要将其合理划分为小节,使用`

    `或`

    `,并用`

    `包裹段落。重要句子可以用``强调。 划分小节: 可以划分为:

    背景与挑战

    Tree ensembles are among the most accurate models for tabular data, yet the computation behind a single prediction is spread over hundreds of decision rules. Feature attributions and counterfactuals explain such predictions through the influence of features or a change of the input, leaving open which of the thresholds applied by the trees a prediction responds to, while structural explanations compress these rules and lose the trees behind them.

    方法:局部决策谓词图与关键谓词

    To address this gap, we propose a local Decision Predicate Graph (local DPG), which gathers the rules executed for a sample into a verifiable record of the computation, with a class support that reproduces every explained prediction. From these rules we select the critical predicate, the executed predicate whose crossing is estimated to move the model most strongly toward a chosen class, which offers the user a concrete threshold and the trees that share it, without an external explainer.

    实验结果

    Across 16 datasets and four ensemble families, it separates differently predicted neighbours better than random and nearest-threshold predicates. A single crossing recovers the true class in 46-69% of wrong predictions, against 29-50% when the same rule is applied to the feature ranked first by TreeSHAP, a measure close to the selection criterion, and on planted faults it finds the faulty threshold in 46-71% of the resulting errors, against 6-21% at random.

    结论

    Although training-free edits of these predicates did not reduce the error on unseen data, the critical predicate complements attributions and counterfactuals by naming where a prediction can be changed.

    检查规则: 1. 只输出HTML片段本身,无Markdown,无解释。 -> 满足 2. 每个自然段落用`

    `,长文用`

    `/`

    `。 -> 满足 3. 列表、引用、代码、重要句子用对应标签。 -> 重要句子用了``,没有列表引用代码。 4. 保留原文所有信息,不得增删改内容。 -> 检查文本是否完全一致。 原文: Tree ensembles are among the most accurate models for tabular data, yet the computation behind a single prediction is spread over hundreds of decision rules. Feature attributions and counterfactuals explain such predictions through the influence of features or a change of the input, leaving open which of the thresholds applied by the trees a prediction responds to, while structural explanations compress these rules and lose the trees behind them. To address this gap, we propose a local Decision Predicate Graph (local DPG), which gathers the rules executed for a sample into a verifiable record of the computation, with a class support that reproduces every explained prediction. From these rules we select the critical predicate, the executed predicate whose crossing is estimated to move the model most strongly toward a chosen class, which offers the user a concrete threshold and the trees that share it, without an external explainer. Across 16 datasets and four ensemble families, it separates differently predicted neighbours better than random and nearest-threshold predicates. A single crossing recovers the true class in 46-69% of wrong predictions, against 29-50% when the same rule is applied to the feature ranked first by TreeSHAP, a measure close to the selection criterion, and on planted faults it finds the faulty threshold in 46-71% of the resulting errors, against 6-21% at random. Although training-free edits of these predicates did not reduce the error on unseen data, the critical predicate complements attributions and counterfactuals by naming where a prediction can be changed. 我拆分的文本:

    Tree ensembles are among the most accurate models for tabular data, yet the computation behind a single prediction is spread over hundreds of decision rules. Feature attributions and counterfactuals explain such predictions through the influence of features or a change of the input, leaving open which of the thresholds applied by the trees a prediction responds to, while structural explanations compress these rules and lose the trees behind them.

    To address this gap, we propose a local Decision Predicate Graph (local DPG), which gathers the rules executed for a sample into a verifiable record of the computation, with a class support that reproduces every explained prediction. From these rules we select the critical predicate, the executed predicate whose crossing is estimated to move the model most strongly toward a chosen class, which offers the user a concrete threshold and the trees that share it, without an external explainer.

    Across 16 datasets and four ensemble families, it separates differently predicted neighbours better than random and nearest-threshold predicates. A single crossing recovers the true class in 46-69% of wrong predictions, against 29-50% when the same rule is applied to the feature ranked first by TreeSHAP, a measure close to the selection criterion, and on planted faults it finds the faulty threshold in 46-71% of the resulting errors, against 6-21% at random.

    Although training-free edits of these predicates did not reduce the error on unseen data, the critical predicate complements attributions and counterfactuals by naming where a prediction can be changed.

    完全一致,没有增删改内容。小标题是新增的,但这是排版要求(长文用h2/h3划分小节),小标题本身不是正文内容,通常可以接受,或者我不加小标题直接用h2作为段落前缀? 规则说:“长文用

    /

    划分小节”,这意味着我可以添加小节标题来划分。为了安全起见,我会添加语义化的中文或英文小标题,因为原文是英文,用英文小标题更合适,或者不加标题直接用h2包裹段落? “长文用

    /

    划分小节”通常意味着添加标题。我添加简洁的英文标题:

    Background and Challenge

    Proposed Method

    Experimental Results

    Conclusion

    等等,规则说“不得增删改内容”,添加标题算不算“增”内容? 为了严格遵循“不得增删改内容”,