Posted on Categories:CS代写, Machine Learning, 机器学习, 计算机代写

机器学习代考_Machine Learning代考_KIT315 Local to Global Hierarchy of Explanation Theories

avatest™

avatest™帮您通过考试

avatest™的各个学科专家已帮了学生顺利通过达上千场考试。我们保证您快速准时完成各时长和类型的考试，包括in class、take home、online、proctor。写手整理各样的资源来或按照您学校的资料教您，创造模拟试题，提供所有的问题例子，以保证您在真实考试中取得的通过率是85%以上。如果您有即将到来的每周、季考、期中或期末考试，我们都能帮助您！

•最快12小时交付

•200+ 英语母语导师

•70分以下全额退款

机器学习代考_Machine Learning代考_Local to Global Hierarchy of Explanation Theories

GLocalX method hierarchically merges local explanations into a global explanation theory. In particular, GLocalX takes as input a set of local explanations in the form of explanation theories $\mathrm{E}={\mathrm{E} 1, \ldots, \mathrm{En}}$ where each theory $\mathrm{Ei}={\mathrm{ei}}$ is formed by a single explanation (i.e., $|\mathrm{Ei}|=1 \forall \mathrm{Ei} \in \mathrm{E}$ ). GLocalX iteratively merges the explanation theories and returns an explanation theory, $\mathrm{E}={\mathrm{e} 01, \ldots, \mathrm{e} 0 \mathrm{k}}$. It emulates the global behavior of the black-box classifier b, simultaneously maintaining the overall simple and interpretable model. At each iteration, GLocalX merges the closest pair of explanation theories $\mathrm{Ei}$, Ej by using a notion of similarity between logical theories. The pairs are filtered out according to the merge quality criterion: if no pair satisfying the criterion is found, GLocalX halts prematurely without building the full hierarchy. The resulting hierarchy of explanation theories can be represented by using a tree-like diagram called a dendrogram. There are two key elements in the GLocalX approach: (i) similarity search, which allows to select which theories to merge and refine, and (ii) a merge function, which allows to refine the explanations.

Given a set of explanation theories $\mathrm{E}$, a pairwise similarity function and a quality criterion, the logical theories are sorted by similarity in a queue Q1. Then, a batch of data is sampled to merge the candidate theories. Using batches instead of the whole training data set favors diverse merges, as the merge procedure has different behaviors according to the data at hand. In the merge loop, the queue is popped to find the most similar pair of theories whose merge satisfies a quality criterion and, the merge operation is run. If the merger is advantageous, the merged theory is kept.

As a quality criterion, the Bayesian information criterion (BIC) is used because it rewards models for their simplicity and accuracy. BIC has been successfully adopted in various techniques, such as clustering or adopting bisecting hierarchical refinement of the model. After a successful merge of the two explanations. Ei, Ej are replaced with the merged theory Ei+j. If no advantageous merge is found, GLocalX halts. This process is iterated until no more merges are possible. Finally, explanations with low fidelity are filtered out to reduce the output size. A $\alpha$ parameter indicates this per-class trimming threshold.

机器学习代考_Machine Learning代考_Finding Similar Theories

Selecting pairs of theories to merge requires the definition of a pairwise similarity function on logical explanation theories. To this aim, the similarity of two theories, E1, E2 is defined as the Jaccard similarity of their coverage on a given instance set $X$.
$$\operatorname{similarity}X\left(E_i, E_j\right)=\frac{\mid \text { coverage }\left(E_i, X\right) \cap \operatorname{coverage}\left(E_j, X\right) \mid}{\mid \text { coverage }\left(E_i, X\right) \cup \operatorname{coverage}\left(E_j, X\right) \mid} \text {. }$$ The larger is the shared coverage of $E_i$ and $E_j$ on $X$, the more similar the two logical explanation theories are. Coverage similarity is a two-faceted measure that captures the premise similarity and the coverage similarity. The former is straightforward: rules with similar premises have similar coverage. The latter balances the premise similarity to avoid rules with similar premises, but low coverage sway the similarity score. The merge function allows GLocalX to generalize a set of explanation theories while balancing fidelity and complexity through approximate logical entailment. The merge involves two operators-join and cut-to simultaneously generalize and preserve a high level of fidelity. In the logical domain, generalization seldom involves premises relaxation or outright removal. Thus, GLocalX advances the state-of-the-art in exploiting this kind of generalization. Pushing for generalization may pull down fidelity. These contrasting behaviors are the focal point in a local to global setting and must be dealt with accordingly. This problem is tackled with a merge function that handles both rule generalization and fidelity. Suppose there are two explanation theories, E1 $=\left{e_1, e_2\right}$ and E2 $=\left{e_1^{\prime}, e_2^{\prime}\right}$, with $e_1, e_1^{\prime}$ explaining a record x1 and $e_2, e_2^{\prime}$ explaining a record x2. \begin{aligned} E_1=\left{e_1\right.&={\text { age } \geq 25, \text { job }=\text { unemployed }, \text { amount } \leq 10 k} \rightarrow \text { deny } \ e_2 &={\text { age } \geq 50, \text { job }=\text { office clerk }} \rightarrow \text { deny }} \ E_2=\left{e_1^{\prime}={\text { age } \geq 20, \text { job }\right.&=\text { manager }, \text { amount }>8 k} \rightarrow \text { accept } \ e_2^{\prime} &={\text { age } \geq 40, \text { job }=\text { office clerk, amount }>5 k} \rightarrow \text { deny }} \end{aligned} The resulting merge yields the following rules $E{1+2}$ :
\begin{aligned} E_{1+2}=\left{e_1^{\prime \prime}\right.&={\text { age } \geq 25, \text { job }=\text { unemployed, amount } \leq 10 k} \rightarrow \text { deny } \ e_2^{\prime \prime} &={\text { age } \in[20,25], \text { job }=\text { manager, amount } \in[8 k, 10 k]} \rightarrow \text { accept } \ e_3^{\prime \prime} &={\text { age } \geq 40} \rightarrow \text { deny }} \end{aligned}

机器学习代考_Machine Learning代考_Local to Global Hierarchy of Explanation Theories

$\mathrm{E}=\mathrm{E} 1, \ldots, \mathrm{En}$ 每个理论在哪里Ei $=$ ei 由单一解释形成 (即, $|\mathrm{Ei}|=1 \forall \mathrm{Ei} \in \mathrm{E}$ ). GLocalx 迭代合并解㸷哩论并返回一个解 论。使用批次而不是整个训练数据集有利于多样化的合并，因为合并过程根据手头的数据具有不同的行为。在合并砶坏中，弹出队 列以找到合并满足质量标准的最相似理论对, 然后运行合并操作。如果合并有利，则保留合并后的理论。 型的二等分层次细化。两种解释成功合并后。Ei、 Ej 被合并后的理论 Ei+j 代替。如果没有找到有利的合并，GLocalX 就会停止。 迭代此过程，直到无法再合并为止。最后，过滤掉保真度低的解释以揻少输出大小。个 $\alpha$ 参数指示此每类修整诃值。

机器学习代考_Machine Learning代考_Finding Similar Theories

$$\text { similarity } X\left(E_i, E_j\right)=\frac{\mid \text { coverage }\left(E_i, X\right) \cap \text { coverage }\left(E_j, X\right) \mid}{\mid \text { coverage }\left(E_i, X\right) \cup \text { coverage }\left(E_j, X\right) \mid} .$$

〈left 缺少或无法识别的分隔符

〈left 缺少或无法识别的分隔符

MATLAB代写

MATLAB 是一种用于技术计算的高性能语言。它将计算、可视化和编程集成在一个易于使用的环境中，其中问题和解决方案以熟悉的数学符号表示。典型用途包括：数学和计算算法开发建模、仿真和原型制作数据分析、探索和可视化科学和工程图形应用程序开发，包括图形用户界面构建MATLAB 是一个交互式系统，其基本数据元素是一个不需要维度的数组。这使您可以解决许多技术计算问题，尤其是那些具有矩阵和向量公式的问题，而只需用 C 或 Fortran 等标量非交互式语言编写程序所需的时间的一小部分。MATLAB 名称代表矩阵实验室。MATLAB 最初的编写目的是提供对由 LINPACK 和 EISPACK 项目开发的矩阵软件的轻松访问，这两个项目共同代表了矩阵计算软件的最新技术。MATLAB 经过多年的发展，得到了许多用户的投入。在大学环境中，它是数学、工程和科学入门和高级课程的标准教学工具。在工业领域，MATLAB 是高效研究、开发和分析的首选工具。MATLAB 具有一系列称为工具箱的特定于应用程序的解决方案。对于大多数 MATLAB 用户来说非常重要，工具箱允许您学习应用专业技术。工具箱是 MATLAB 函数（M 文件）的综合集合，可扩展 MATLAB 环境以解决特定类别的问题。可用工具箱的领域包括信号处理、控制系统、神经网络、模糊逻辑、小波、仿真等。