注册    登录
创作新主题
社区所有版块导航
Python
python开源   Django   Python   DjangoApp   pycharm  
DATA
docker   Elasticsearch  
分享
问与答   闲聊   招聘   翻译   创业   分享发现   分享创造   求职   区块链   支付之战  
aigc
aigc   chatgpt  
WEB开发
linux   MongoDB   Redis   DATABASE   NGINX   其他Web框架   web工具   zookeeper   tornado   NoSql   Bootstrap   js   peewee   Git   bottle   IE   MQ   Jquery  
机器学习
机器学习算法  
Python88.com
反馈   公告   社区推广  
产品
短视频  
印度
印度  
关注
Py学习  »  机器学习算法

TikZ 复现深度学习五大算法图 - 经典算法全打尽

LaTeX工作室 • 6 月前 • 194 次点击  

TikZ 复现深度学习五大算法图

本文分享通过 TikZ 绘制深度学习五大算法图，其中深度学习算法内容来源于米酥，以下介绍仅部分算法给出图的完整代码。

全部的源代码打包下载地址为TikZ 复现深度学习五大算法图https://www.latexstudio.net/index/details/index/mid/4132.html。

1. 卷积神经网络（CNN）

卷积神经网络(Convolutional Neural Networks, CNN)：主要用于图像识别和处理任务，通过卷积层、池化层和全连接层来提取图像特征和进行分类(由于篇幅问题，本文未展示全部代码)。

\documentclass[tikz,border=10pt]{standalone}
\usepackage{ctex,bm,xcolor}
\usetikzlibrary{calc, positioning, arrows.meta, shapes}
\newcommand{\vars}[2]{${\bm#1}_{\bm#2}$}
\usepackage{fontspec}
\usepackage{fontenc}
\setmainfont{Times New Roman}

\tikzset{
......
}

\begin{document}

\begin{tikzpicture}

%%源代码请在链接地址下载

\end{tikzpicture}
\end{document}

2. 递归神经网络（RNN）

递归神经网络 (Recurrent Neural Networks,RNN)：用于处理序列数据，具有记忆能力，可以在自然语言处理、语音识别等领域发挥重要作用(完整代码)。

\documentclass[tikz,border=10pt]{standalone}
\usepackage{tikz-network}
\usetikzlibrary{calc, positioning, arrows.meta, shapes}
\usepackage{fontspec}
\usepackage{fontenc}
\setmainfont{Times New Roman}

\begin{document}
\sffamily
\begin{tikzpicture}
\Vertex[label=$i_1$,style={line width=1pt,fill=green!45}]{A1}
\Vertex[y=1,label=$i_0$,style={line width=1pt,fill=green!45}]{A2}

\foreach \y/\n in {1.5/1,0.5/2,-0.5/3}{
\Vertex[y=\y,x=2.5,label=$h_\n$,style={line width=1pt,fill=yellow!45}]{B\n}
}

\foreach \y/\n in {-1.5/4,-2.5/5,-3.5/6}{
\Vertex[y=\y,x=2.5,label=$h_\n$,style={line width=1pt,fill=violet!45}]{B\n}
}

\Vertex[x=5,label=$O_1$,style={line width=1pt,fill=cyan!15}]{C1}
\Vertex[x=5,y=1,label=$O_0$,style={line width=1pt,fill=cyan!15}]{C2}

\foreach \i in {1,2}{
\foreach \n in {1,2,3}{
\Edge[Direct,style={line width=0.75pt,black}](A\i)(B\n)
}}

\foreach \i in {1,2,3}{
\foreach \n in {1,2}{
\Edge[Direct,style={line width=0.75pt,black}](B\i)(C\n)
}}

\Edge[Direct,style={line width=0.75pt,black},bend=45](B4)(B1)
\Edge[Direct,style={line width=0.75pt,black},bend=45](B5)(B2)
\Edge[Direct,style={line width=0.75pt,black},bend=45](B6)(B3)

\Edge[Direct,style={line width=0.75pt,black},bend=45](B1)(B4)
\Edge[Direct,style={line width=0.75pt,black},bend=45](B2)(B5)
\Edge[Direct,style={line width=0.75pt,black},bend=45](B3)(B6)

\node[above=1cm] at (A2){Input Layer};
\node[above=0.5cm] at (B1){Hidden Layer};
\node[above=1cm] at (C2){Output Layer};

\end{tikzpicture}
\end{document}

3. 长短期记忆网络（LSTM）

长短期记忆网络 (Long Short-Term Memory,LSTM)：RNN 的一种特殊形式，用于解决传统 RNN 在处理长序列数据时出现的梯度消失或梯度爆炸的问题，适用于时间序列预测等任务(完整代码)。

\documentclass[tikz,border=10pt]{standalone}



    
\usepackage{ctex,bm,xcolor}
\usetikzlibrary{calc, positioning, arrows.meta, shapes}
\newcommand{\vars}[2]{${\bm#1}_{\bm#2}$}
\usepackage{fontspec}
\usepackage{fontenc}
\setmainfont{Times New Roman}

\tikzset{
    myfont/.style={
        font=\footnotesize\sffamily
    },
    frame/.style={
        rectangle,
        rounded corners=5mm,
        draw,gray,
        fill=green!15!white,
        very thick,
    },
    add/.style={
        circle,
        draw, thick,
        fill=yellow!60,
        inner sep=.5pt,
        minimum height =3mm,
    },
    prod/.style={
        circle,
        draw, thick,
        fill=lightgray!40!blue!30,
        inner sep=.5pt,
        minimum height =3mm,
    },
    function/.style={
        ellipse,
        draw, thick,
        fill=cyan!30,
        inner sep=1pt
    },
    cell/.style={
        circle,draw,fill=magenta!30,
        line width = .75pt,
        minimum width=8mm,
        inner sep=1pt,
    },
    hidden/.style={
        circle,draw,fill=orange!50,
        line width = .75pt,
        minimum width=8mm,
        inner sep=1pt,
    },
    input/.style={
        circle,draw,
        fill=teal!30,
        line width = .75pt,
        minimum width=8mm,
        inner sep=1pt,
    },
    actfunc/.style={
        rectangle, draw,
        fill=pink!30,
        minimum width=4.25mm,
        minimum height=3.75mm,
        inner sep=1pt,
        thick,
    },
    ArrowC1/.style={
        rounded corners=2mm,
        >=Stealth[round],
        very thick,
    },
    ArrowC2/.style={
        rounded corners=3mm,
        >=Stealth[round],
        very thick,
    }
}

\begin{document}

\begin{tikzpicture}

%%% First ===========================================
\begin{scope}[xshift=0em,yshift=0em]

\node [frame, minimum height =4cm, minimum width=6cm] at (0,0) {};
    \node  (ibox1) at (-2,-0.8) {};
    \node  (ibox2) at (-1.35,-0.8) {};
    \node  (ibox3) at (-0.45,-0.8) {};
    \node  (ibox4) at (0.5,-0.8) {};

    \node (mux1) at (-2,1.5) {};
    \node (add1) at (-0.5,1.5) {};
    \node (mux2) at (-0.5,0) {};
    \node  (mux3) at (1.5,0) {};
    \node  (func1) at (1.5,0.75) {};

    \node(c) at (-3.1,1.5) {};
    \node(h) at (-3.1,-1.5) {};
    \node[input] (x) at (-2.5,-3) {\vars{x}{t-1}};

   \node (c2) at (4,1.5) {};
   \node (h2) at (4,-1.5) {};
    \node[hidden,fill=violet!45] (x2) at (2.5,3) {\vars{h}{t-1}};
\draw [->, ArrowC1] (c) -- (mux1) -- (add1) -- (c2);

    \draw [opacity=0.3,ArrowC2] (h) -| (ibox4);
    \draw [opacity=0.3,ArrowC1] (h) -| (ibox1);
    \draw [opacity=0.3,ArrowC1] (h) -| (ibox2);
    \draw [opacity=0.3,ArrowC1] (h) -| (ibox3);
    \draw [ArrowC1] (x) -- (x |- h) -| (ibox1);

    \draw [->, opacity=0.3,ArrowC2] (ibox1) -- (mux1);
    \node [right=1.5pt] () at ($(ibox1)!.5!(mux1)$) {};
    \draw [->,opacity=0.3, ArrowC2] (ibox2) |- (mux2);
    \node [right=2pt] () at (ibox2 |- mux2) {};
    \draw [->, opacity=0.3,ArrowC2] (ibox3) -- (mux2);
    \node [label=right:{\sffamily\Huge A}, left=.5pt] () at ($(ibox3)!.5!(mux2)$) {};
    \draw [->,opacity=0.3, ArrowC2] (ibox4) |- (mux3);
    \node [right=2pt] () at (ibox4 |- mux3) {};
    \draw [->,opacity=0.3, ArrowC2] (mux2) -- (add1);
    \draw [->, opacity=0.3,ArrowC1] (add1) -| (func1);
    \draw [->,opacity=0.3, ArrowC2] (func1) -- (mux3);

    \draw [->, ArrowC2] (mux3) |- (h2);
    \draw (c2 -| x2) ++(0,-0.15) coordinate (i1);
    \draw [-, ArrowC2] (h2 -| x2) -| (i1);
    \draw [->, ArrowC2] (i1)++(0,0.3) -- (x2);
\end{scope}


%%% Second ===========================================
\begin{scope}[xshift=18.5em,yshift=0em]
\node [frame, minimum height =4cm, minimum width=6cm] at (0,0) {};
    \node [actfunc] (ibox1) at (-2,-0.8) {$\bm\sigma$};
    \node [actfunc] (ibox2) at (-1.35,-0.8) {$\bm\sigma$};
    \node [actfunc, minimum width=9mm] (ibox3) at (-0.45,-0.8) {$\bm\tanh$};
    \node [actfunc] (ibox4) at (0.5,-0.8) {$\bm\sigma$};

    \node [prod] (mux1) at (-2,1.5) {$\bm\times$};
    \node [add] (add1) at (-0.5,1.5) {$\bm+$};
    \node [prod] (mux2) at (-0.5,0) {$\bm\times$};
    \node [prod] (mux3) at (1.5,0) {$\bm\times$};
    \node [function] (func1) at (1.5,0.75) {$\bm\tanh$};

%    \node[cell] (c) at (-4,1.5) {\vars{c}{t-1}};
%   \node[hidden] (h) at (-4,-1.5) {\vars{h}{t-1}};
    \node[input] (x) at (-2.5,-3) {\vars{x}{t}};

    \node (c2) at (4,1.5) {};
    \node(h2) at (4,-1.5) {};
    \node[hidden] (x2) at (2.5,3) {\vars{h}{t}};
\draw [->, ArrowC1] (c) -- (mux1) -- (add1) -- (c2);

    \draw [ArrowC2] (h) -| (ibox4);
    \draw [ArrowC1] (h) -| (ibox1);
    \draw [ArrowC1] (h) -| (ibox2);
    \draw [ArrowC1] (h) -| (ibox3);
    \draw [ArrowC1] (x) -- (x |- h) -| (ibox1);

    \draw [->, ArrowC2] (ibox1) -- (mux1);
    \node [label=left:$\bm{f_t}$,right=1.5pt] () at ($(ibox1)!.5!(mux1)$) {};
    \draw [->, ArrowC2] (ibox2) |- (mux2);
    \node [label=left:$\bm{i_t}$, right=2pt] () at (ibox2 |- mux2) {};
    \draw [->, ArrowC2] (ibox3) -- (mux2);
    \node [label=right:$\bm{\tilde{C}_t}$, left=.5pt] () at ($(ibox3)!.5!(mux2)$) {};
    \draw [->, ArrowC2] (ibox4) |- (mux3);
    \node [label=left:$\bm{o_t}$, right=2pt] () at (ibox4 |- mux3) {};
    \draw [->, ArrowC2] (mux2) -- (add1);
    \draw [->, ArrowC1] (add1) -| (func1);
    \draw [->, ArrowC2] (func1) -- (mux3);

    \draw [->, ArrowC2] (mux3) |- (h2);
    \draw (c2 -| x2) ++(0,-0.15) coordinate (i1);
    \draw [-, ArrowC2] (h2 -| x2) -| (i1);
    \draw [->, ArrowC2] (i1)++(0,0.3) -- (x2);
\end{scope}

%%% third ===========================================
\begin{scope}[xshift=37em,yshift=0em]

\node [frame, minimum height =4cm, minimum width=6cm] at (0,0) {};
    \node  (ibox1) at (-2,-0.8) {};
    \node  (ibox2) at (-1.35,-0.8) {};
    \node  (ibox3) at (-0.45,-0.8) {};
    \node  (ibox4) at (0.5,-0.8) {};

    \node (mux1) at (-2,1.5) {};
    \node (add1) at (-0.5,1.5) {};
    \node (mux2) at (-0.5,0) {};
    \node  (mux3) at (1.5,0) {};
    \node  (func1) at (1.5,0.75) {};

    \node(c) at (-3.1,1.5) {};
    \node(h) at (-3.1,-1.5) {};
    \node[input] (x) at (-2.5,-3) {\vars{x}{t+1}};

   \node (c2) at (4,1.5) {};
   \node (h2) at (4,-1.5) {};
    \node[hidden,fill=violet!45] (x2) at (2.5,3) {\vars{h}{t+1}};
\draw [->, ArrowC1] (c) -- (mux1) -- (add1) -- (c2);

    \draw [opacity=0.3,ArrowC2] (h) -| (ibox4);
    \draw [opacity=0.3,ArrowC1] (h) -| (ibox1);
    \draw [opacity=0.3,ArrowC1] (h) -| (ibox2);
    \draw [opacity=0.3,ArrowC1] (h) -| (ibox3);
    \draw [ArrowC1] (x) -- (x |- h) -| (ibox1);

    \draw [->, opacity=0.3,ArrowC2] (ibox1) -- (mux1);
    \node [right=1.5pt] () at ($(ibox1)!.5!(mux1)$) {};
    \draw [->,opacity=0.3, ArrowC2] (ibox2) |- (mux2);
    \node [right=2pt] () at (ibox2 |- mux2) {};
    \draw [->, opacity=0.3,ArrowC2] (ibox3) -- (mux2);
    \node [label=right:{\sffamily\Huge A}, left=.5pt] () at ($(ibox3)!.5!(mux2)$) {};
    \draw [->,opacity=0.3, ArrowC2] (ibox4) |- (mux3);
    \node [right=2pt] () at (ibox4 |- mux3) {};
    \draw [->,opacity=0.3, ArrowC2] (mux2) -- (add1);
    \draw [->, opacity=0.3,ArrowC1] (add1) -| (func1);
    \draw [->,opacity=0.3, ArrowC2] (func1) -- (mux3);

    \draw [->, ArrowC2] (mux3) |- (h2);
    \draw (c2 -| x2) ++(0,-0.15) coordinate (i1);
    \draw [-, ArrowC2] (h2 -| x2) -| (i1);
    \draw [->, ArrowC2] (i1)++(0,0.3) -- (x2);
\end{scope}

\end{tikzpicture}
\end{document}

4. 生成对抗网络（GAN）

生成对抗网络 (Generative Adversarial Networks, GAN)：由生成器和判别器组成的框架，用于生成逼真的样本数据，被广泛应用于图像生成、视频生成等领域（源代码请到网站下载）。

5. 强化学习（RL）

强化学习 (Reinforcement Learning, RL)：以智能体与环境的交互来达到特定目标的一种机器学习方式。其中，深度 Q 网络 (Deep Q-Network.DQN)是一种融合深度学习和强化学习的方法，被用于解决复杂的決策问题（源代码请到网站下载）。

最后，感谢大家的支持和关注！

关注我们公众号，学习 LaTeX 无烦恼！

入门资料，免费知识代码：

https://flowus.cn/latex/share/66110e84-b24a-4cd5-b8a7-2ba2afb35a30

精心制作免费视频教程：

https://space.bilibili.com/209746320

Python社区是高质量的Python/Django开发社区
本文地址：http://www.python88.com/topic/172322
Reply
 
194 次点击  
登录后回复
关于   移动版
Py学习 - 专注于Python技术发展的社区(原Django社区)
沪ICP备11025650号