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

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

The multi-head self-attention mechanism is the core part of the Transformer block. Based on every time step of the input sequence $\left(x_1^{\prime}, x_2^{\prime}, \ldots\right)$, a query, a key, and a value are calculated through a matrix multiplication with a set of three learnable weight matrices, implemented as fully connected layers. The output after each FC layer yields a $Q, K$, and $V$ matrix that contains the query, key, and value vector of length $d_{\mathrm{tf}}$ at each time step. Instead of performing a single attention function with $d_{\mathrm{tf}}$-dimensional queries, keys, and values, the attention function is performed in parallel with multiple “heads”. The multi-head mechanism allows the model to jointly attend to information from different representation subspaces at different positions. The outputs of each head are concatenated and multiplied with a weight matrix (implemented as FC layer) to output a sequence of updated features with dimension $d_{\mathrm{tf}}$.

## 计算机代写|深度学习代写Deep Learning代考|Dot-Product Self-Attention

The attention mechanism can be described as mapping a query and a set of keyvalue pairs to an output. The output is computed as a weighted sum of the values, where the weight assigned to each value is computed by a compatibility function of the query with the corresponding key. In the case of a self-attention mechanism, the sequence of queries, keys, and values are produced by three different FC layers that all use the same input sequence $x^{\prime}$. The resulting matrices $Q, K$, and $V$ are the input for the attention mechanism, where the queries and keys are compared to each other through the dot product. To obtain the dot product for all time steps at once, a matrix multiplication (MatMul) between $Q$ and $K$ is calculated. Thus, the query of each time step is compared to the keys of all time steps in the sequences, yielding an $L \times L$ attention score matrix, where $L$ is the sequence length. Because for large dimensions $d_{\mathrm{tf}}$, the gradient can become extremely small, the score matrix is divided by $d_{\mathrm{tf}}$ (i.e. scaling). A softmax function is then applied to normalise the attention scores and to obtain the weights on the values. The resulting weights are finally applied to the value matrix $V$ through another matrix multiplication. Thus, the output is computed as
$$\operatorname{Attention}(Q, K, V)=\operatorname{softmax}\left(\frac{Q K^T}{\sqrt{d_k}}\right) V$$
As a result, each time step in the sequence is made up of a weighted average of all sequence steps. In this way, each time step can add information from other time steps in the sequence.

Because in this work speech signals of variable length are used as inputs, the feature sequences need to be zero-padded to the length $L$ of the longest sequence before passing them to the self-attention network. To avoid that the network attends to these zero-padded time steps, a mask is applied to the attention score $L \times L$ matrix. All values in the input of the softmax function that correspond to zeropadded time steps are masked out by setting them to $-\infty$.

## 计算机代写|深度学习代写Deep Learning代考|Dot-Product SelfAttention

$$\operatorname{Attention}(Q, K, V)=\operatorname{softmax}\left(\frac{Q K^T}{\sqrt{d_k}}\right) V$$

## 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。