<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta charset="utf-8">
<meta name="generator" content="pdf2htmlEX">
<meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1">
<link rel="stylesheet" href="https://static.pudn.com/base/css/base.min.css">
<link rel="stylesheet" href="https://static.pudn.com/base/css/fancy.min.css">
<link rel="stylesheet" href="https://static.pudn.com/prod/directory_preview_static/62466b0a54457c44d0878d59/raw.css">
<script src="https://static.pudn.com/base/js/compatibility.min.js"></script>
<script src="https://static.pudn.com/base/js/pdf2htmlEX.min.js"></script>
<script>
try{
pdf2htmlEX.defaultViewer = new pdf2htmlEX.Viewer({});
}catch(e){}
</script>
<title></title>
</head>
<body>
<div id="sidebar" style="display: none">
<div id="outline">
</div>
</div>
<div id="pf1" class="pf w0 h0" data-page-no="1"><div class="pc pc1 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/62466b0a54457c44d0878d59/bg1.jpg"><div class="c x1 y1 w2 h2"><div class="t m0 x2 h3 y2 ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m0 x2 h3 y3 ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m0 x2 h3 y4 ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m0 x2 h3 y5 ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m0 x2 h3 y6 ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m0 x3 h3 y7 ff1 fs0 fc0 sc0 ls1 ws0">Neural Net<span class="_ _0"></span>works and <span class="_ _0"></span>Deep Learn<span class="_ _0"></span>ing<span class="ls0"> </span></div><div class="t m0 x2 h3 y8 ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m0 x4 h4 y9 ff2 fs0 fc0 sc0 ls2 ws0"><span class="ff1 fc1 ls3">Michael<span class="_ _0"></span> Niels<span class="_ _0"></span>en<span class="fc0 ls0"> </span></span></div><div class="t m0 x5 h5 ya ff3 fs1 fc0 sc0 ls0 ws0"></div><div class="t m0 x5 h5 yb ff3 fs1 fc0 sc0 ls0 ws0"></div><div class="t m0 x6 h5 yc ff3 fs1 fc0 sc0 ls4 ws0"><span class="ls0"><span class="fc2 ls5"><span class="_ _0"></span><span class="_ _0"></span><span class="_ _0"></span><span class="fc0 ls0"></span></span></span></div><div class="t m0 x2 h5 yd ff3 fs1 fc0 sc0 ls0 ws0"></div><div class="t m0 x7 h5 ye ff3 fs1 fc0 sc0 ls6 ws0"><span class="_ _1"> </span><span class="ls4"><span class="ls0"></span></span></div><div class="t m0 x2 h5 yf ff3 fs1 fc0 sc0 ls0 ws0"></div><div class="t m0 x8 h5 y10 ff3 fs1 fc0 sc0 ls4 ws0"><span class="fc2 ls5"><span class="_ _0"></span><span class="ls7"><span class="fc0 ls0"></span></span></span></div><div class="t m0 x5 h6 y11 ff3 fs2 fc0 sc0 ls0 ws0"></div><div class="t m0 x5 h6 y12 ff3 fs2 fc0 sc0 ls0 ws0"></div><div class="t m0 x9 h7 y13 ff4 fs3 fc0 sc0 ls8 ws0"><span class="ls0"></span></div><div class="t m0 x5 h6 y14 ff3 fs2 fc0 sc0 ls0 ws0"><span class="ls9"></span><span class="lsa"></span></div><div class="t m0 x5 h6 y15 ff3 fs2 fc0 sc0 lsb ws0"><span class="ls0"><span class="_ _2"> </span><span class="lsc"><span class="_ _2"> </span></span></span></div><div class="t m0 x5 h6 y16 ff3 fs2 fc0 sc0 ls0 ws0"><span class="_ _2"> </span><span class="lsc"><span class="_ _2"> </span></span></div><div class="t m0 x5 h6 y17 ff3 fs2 fc0 sc0 ls0 ws0"></div><div class="t m0 x5 h6 y18 ff3 fs2 fc0 sc0 ls0 ws0"><span class="_ _2"> </span><span class="lsc"><span class="_ _2"> </span></span></div><div class="t m0 x5 h6 y19 ff3 fs2 fc0 sc0 ls0 ws0"></div></div></div><div class="pi" data-data='{"ctm":[1.611639,0.000000,0.000000,1.611639,0.000000,0.000000]}'></div></div>
</body>
</html>
<div id="pf2" class="pf w0 h0" data-page-no="2"><div class="pc pc2 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/62466b0a54457c44d0878d59/bg2.jpg"><div class="c xa y1 w3 h8"><div class="t m0 xb h9 y1a ff5 fs4 fc0 sc0 ls0 ws0">2016/10/10<span class="_ _3"> </span>Neural networks and deep learning</div><div class="t m0 xb h9 y1b ff5 fs4 fc0 sc0 ls0 ws0">http://neuralnetworksanddeeplearning.com/index.html<span class="_ _4"> </span>1/2</div></div><div class="c x5 y1c w4 ha"><div class="t m0 xc hb y1d ff6 fs5 fc3 sc0 ls0 ws0">NeuralNetworksandDeepLearning<span class="ff7"> is a free online book. The</span></div><div class="t m0 xc hb y1e ff7 fs5 fc3 sc0 ls0 ws0">book will teach you about:</div><div class="t m0 xd hb y1f ff7 fs5 fc0 sc0 ls0 ws0">Neural networks, a beautiful biologically-inspired</div><div class="t m0 xd hb y20 ff7 fs5 fc0 sc0 ls0 ws0">programming paradigm which enables a computer to learn</div><div class="t m0 xd hb y21 ff7 fs5 fc0 sc0 ls0 ws0">from observational data</div><div class="t m0 xd hb y22 ff7 fs5 fc0 sc0 ls0 ws0">Deep learning, a powerful set of techniques for learning in</div><div class="t m0 xd hb y23 ff7 fs5 fc0 sc0 ls0 ws0">neural networks</div><div class="t m0 xc hb y24 ff7 fs5 fc0 sc0 ls0 ws0">Neural networks and deep learning currently provide the best</div><div class="t m0 xc hb y25 ff7 fs5 fc0 sc0 ls0 ws0">solutions to many problems in image recognition, speech</div><div class="t m0 xc hb y26 ff7 fs5 fc0 sc0 ls0 ws0">recognition, and natural language processing. This book will teach</div><div class="t m0 xc hb y27 ff7 fs5 fc0 sc0 ls0 ws0">you many of the core concepts behind neural networks and deep</div><div class="t m0 xc hb y28 ff7 fs5 fc0 sc0 ls0 ws0">learning.</div><div class="t m0 xc hb y29 ff7 fs5 fc3 sc0 ls0 ws0">For more details about the approach taken in the book, <span class="fc4">see here</span>. Or</div><div class="t m0 xc hb y2a ff7 fs5 fc3 sc0 ls0 ws0">you can jump directly to <span class="fc4">Chapter 1</span> and get started.</div><div class="t m0 xc hc y2b ff8 fs6 fc5 sc0 ls0 ws0">Neural Networks and Deep Learning</div><div class="t m0 xe hd y2c ff7 fs7 fc4 sc0 ls0 ws0">Neural Networks and Deep earning</div><div class="t m0 xf hd y2d ff7 fs7 fc4 sc0 ls0 ws0">hat this book is about</div><div class="t m0 xf hd y2e ff7 fs7 fc4 sc0 ls0 ws0">On the eercises and problems</div><div class="t m0 xf hd y2f ff7 fs7 fc4 sc0 ls0 ws0">sing neural nets to recognie</div><div class="t m0 xf hd y30 ff7 fs7 fc4 sc0 ls0 ws0">handwritten digits</div><div class="t m0 xf hd y31 ff7 fs7 fc4 sc0 ls0 ws0">ow the backpropagation</div><div class="t m0 xf hd y32 ff7 fs7 fc4 sc0 ls0 ws0">algorithm works</div><div class="t m0 xf hd y33 ff7 fs7 fc4 sc0 ls0 ws0">mproving the way neural</div><div class="t m0 xf hd y34 ff7 fs7 fc4 sc0 ls0 ws0">networks learn</div><div class="t m0 xf hd y35 ff7 fs7 fc4 sc0 ls0 ws0"> visual proof that neural nets can</div><div class="t m0 xf hd y36 ff7 fs7 fc4 sc0 ls0 ws0">compute any function</div><div class="t m0 xf hd y37 ff7 fs7 fc4 sc0 ls0 ws0">hy are deep neural networks</div><div class="t m0 xf hd y38 ff7 fs7 fc4 sc0 ls0 ws0">hard to train</div><div class="t m0 xf hd y39 ff7 fs7 fc4 sc0 ls0 ws0">Deep learning</div><div class="t m0 xf hd y3a ff7 fs7 fc4 sc0 ls0 ws0">ppendi: s there a <span class="ff6">siple</span></div><div class="t m0 xf hd y3b ff7 fs7 fc4 sc0 ls0 ws0">algorithm for intelligence</div><div class="t m0 xf hd y3c ff7 fs7 fc4 sc0 ls0 ws0">cknowledgements</div><div class="t m0 xf hd y3d ff7 fs7 fc4 sc0 ls0 ws0">Frequently sked uestions</div><div class="t m0 xe hd y3e ff7 fs7 fc3 sc0 ls0 ws0">f you benefit from the book, please</div><div class="t m0 xe hd y3f ff7 fs7 fc3 sc0 ls0 ws0">make a small donation.  suggest ,</div><div class="t m0 xe hd y40 ff7 fs7 fc3 sc0 ls0 ws0">but you can choose the amount.</div><div class="t m0 xe he y41 ff9 fs8 fc0 sc0 ls0 ws0">Sponsors</div><div class="t m0 xe hd y42 ff7 fs7 fc3 sc0 ls0 ws0">Thanks to all the <span class="fc4">supporters</span> who</div><div class="t m0 xe hd y43 ff7 fs7 fc3 sc0 ls0 ws0">made the book possible, with</div><div class="t m0 xe hd y44 ff7 fs7 fc3 sc0 ls0 ws0">especial thanks to avel Dudrenov.</div><div class="t m0 xe hd y45 ff7 fs7 fc3 sc0 ls0 ws0">Thanks also to all the contributors to</div><div class="t m0 xe hd y46 ff7 fs7 fc3 sc0 ls0 ws0">the <span class="fc4">ugfinder all of Fame</span>.</div><div class="t m0 xe he y47 ff9 fs8 fc0 sc0 ls0 ws0">Resources</div><div class="t m0 xe hd y48 ff7 fs7 fc4 sc0 ls0 ws0">ook F</div><div class="t m0 xe hd y49 ff7 fs7 fc4 sc0 ls0 ws0">Code repository</div><div class="t m0 xe hd y4a ff7 fs7 fc4 sc0 ls0 ws0">ichael Nielsens project</div><div class="t m0 xe hd y4b ff7 fs7 fc4 sc0 ls0 ws0">announcement mailing list</div><div class="t m0 xe hd y4c ff7 fs7 fc4 sc0 ls0 ws0">Deep earning<span class="fc3">, draft book in</span></div><div class="t m0 xe hd y4d ff7 fs7 fc3 sc0 ls0 ws0">preparation, by oshua engio, an</div><div class="t m0 xe hd y4e ff7 fs7 fc3 sc0 ls0 ws0">oodfellow, and aron Courville</div></div><div class="c x10 y4f w5 hf"><div class="t m0 x0 h10 y50 ffa fs9 fc0 sc0 ls0 ws0">欧拉的博客<span class="ffb">:www.liuhao.me</span></div></div></div><div class="pi" data-data='{"ctm":[1.611639,0.000000,0.000000,1.611639,0.000000,0.000000]}'></div></div>
<div id="pf3" class="pf w0 h0" data-page-no="3"><div class="pc pc3 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/62466b0a54457c44d0878d59/bg3.jpg"><div class="c xa y1 w3 h8"><div class="t m0 xb h9 y1a ff5 fs4 fc0 sc0 ls0 ws0">2016/10/10<span class="_ _3"> </span>Neural networks and deep learning</div><div class="t m0 xb h9 y1b ff5 fs4 fc0 sc0 ls0 ws0">http://neuralnetworksanddeeplearning.com/index.html<span class="_ _4"> </span>2/2</div></div><div class="c x5 y1c w4 ha"><div class="t m0 xe hd y51 ff7 fs7 fc3 sc0 ls0 ws0">y <span class="fc4">ichael Nielsen</span>  an 1</div><div class="t m0 xc h11 y52 ff9 fsa fc5 sc0 ls0 ws0">ncecorpesecesoosceeseneureorsneepernn</div><div class="t m0 xc h11 y53 ff9 fsa fc5 sc0 ls0 ws0">eernonress</div><div class="t m0 xc h11 y54 ff9 fsa fc5 sc0 ls0 ws0">sorscenseunerreeoonsruononoercnporecensesens</div><div class="t m0 xc h11 y55 ff9 fsa fc5 sc0 ls0 ws0">ourereeocop<span class="_ _0"></span>srenuonsoounooseoureneresencoercusepese</div><div class="t m0 xc h11 y56 ff9 fsa fc4 sc0 ls0 ws0">conce<span class="fc5"></span></div><div class="t m0 xf h11 y52 ff9 fsa fc5 sc0 ls0 ws0">supern</div></div><div class="c x10 y4f w5 hf"><div class="t m0 x0 h10 y50 ffa fs9 fc0 sc0 ls0 ws0">欧拉的博客<span class="ffb">:www.liuhao.me</span></div></div></div><div class="pi" data-data='{"ctm":[1.611639,0.000000,0.000000,1.611639,0.000000,0.000000]}'></div></div>
<div id="pf4" class="pf w0 h0" data-page-no="4"><div class="pc pc4 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/62466b0a54457c44d0878d59/bg4.jpg"><div class="c xa y1 w3 h8"><div class="t m0 xb h9 y1a ff5 fs4 fc0 sc0 ls0 ws0">2016/10/10<span class="_ _3"> </span>Neural networks and deep learning</div><div class="t m0 xb h9 y1b ff5 fs4 fc0 sc0 ls0 ws0">http://neuralnetworksanddeeplearning.com/aout.html<span class="_ _4"> </span>1/</div></div><div class="c x5 y1c w4 ha"><div class="t m0 xc hb y1d ff7 fs5 fc3 sc0 ls0 ws0">Neural networks are one of the most beautiful programming</div><div class="t m0 xc hb y1e ff7 fs5 fc3 sc0 ls0 ws0">paradigms ever invented. n the conventional approach to</div><div class="t m0 xc hb y57 ff7 fs5 fc3 sc0 ls0 ws0">programming, we tell the computer what to do, breaking big</div><div class="t m0 xc hb y58 ff7 fs5 fc3 sc0 ls0 ws0">problems up into many small, precisely defined tasks that the</div><div class="t m0 xc hb y59 ff7 fs5 fc3 sc0 ls0 ws0">computer can easily perform. y contrast, in a neural network we</div><div class="t m0 xc hb y5a ff7 fs5 fc3 sc0 ls0 ws0">dont tell the computer how to solve our problem. nstead, it learns</div><div class="t m0 xc hb y5b ff7 fs5 fc3 sc0 ls0 ws0">from observational data, figuring out its own solution to the</div><div class="t m0 xc hb y5c ff7 fs5 fc3 sc0 ls0 ws0">p</div><div class="t m0 x11 hb y5d ff7 fs5 fc3 sc0 ls0 ws0">roblem at hand.</div><div class="t m0 xc hb y5e ff7 fs5 fc3 sc0 ls0 ws0">utomatically learning from data sounds promising. owever, until</div><div class="t m0 xc hb y5f ff7 fs5 fc3 sc0 ls0 ws0"> we didnt know how to train neural networks to surpass more</div><div class="t m0 xc hb y60 ff7 fs5 fc3 sc0 ls0 ws0">traditional approaches, ecept for a few specialied problems. hat</div><div class="t m0 xc hb y61 ff7 fs5 fc3 sc0 ls0 ws0">changed in  was the discovery of techniques for learning in so-</div><div class="t m0 xc hb y62 ff7 fs5 fc3 sc0 ls0 ws0">called deep neural networks. These techniques are now known as</div><div class="t m0 xc hb y63 ff7 fs5 fc3 sc0 ls0 ws0">deep learning. Theyve been developed further, </div><div class="t m0 x12 hb y64 ff7 fs5 fc3 sc0 ls0 ws0">and today deep</div><div class="t m0 xc hb y65 ff7 fs5 fc3 sc0 ls0 ws0">neural networks and deep learning achieve outstanding</div><div class="t m0 xc hb y66 ff7 fs5 fc3 sc0 ls0 ws0">performance on many important problems in computer vision,</div><div class="t m0 xc hb y67 ff7 fs5 fc3 sc0 ls0 ws0">speech recognition, and natural language processing. Theyre being</div><div class="t m0 xc hb y68 ff7 fs5 fc3 sc0 ls0 ws0">deployed on a large scale by companies such as oogle, icrosoft,</div><div class="t m0 xc hb y69 ff7 fs5 fc3 sc0 ls0 ws0">and Facebook.</div><div class="t m0 xc hb y6a ff7 fs5 fc3 sc0 ls0 ws0">The purpose of this book is to help you master the core concepts of</div><div class="t m0 xc hb y6b ff7 fs5 fc3 sc0 ls0 ws0">neural networks, including modern techniques for deep learning.</div><div class="t m0 xc hb y6c ff7 fs5 fc3 sc0 ls0 ws0"></div><div class="t m0 x13 hb y6d ff7 fs5 fc3 sc0 ls0 ws0">fter working through the book you will have written code that uses</div><div class="t m0 xc hb y6e ff7 fs5 fc3 sc0 ls0 ws0">neural networks and deep learning to solve comple pattern</div><div class="t m0 xc hb y6f ff7 fs5 fc3 sc0 ls0 ws0">recognition problems. nd you will have a foundation to use neural</div><div class="t m0 xc hb y70 ff7 fs5 fc3 sc0 ls0 ws0">networks and deep learning to attack problems of your own</div><div class="t m0 xc hb y71 ff7 fs5 fc3 sc0 ls0 ws0">devising.</div><div class="t m0 xc h12 y72 ff8 fsb fc4 sc0 ls0 ws0"> prinipleoriented approa</div><div class="t m0 xc hb y73 ff7 fs5 fc3 sc0 ls0 ws0">One conviction underlying the book is that its better to obtain a</div><div class="t m0 xc hb y74 ff7 fs5 fc3 sc0 ls0 ws0">solid understanding of the core principles of neural networks and</div><div class="t m0 xc hb y75 ff7 fs5 fc3 sc0 ls0 ws0">d</div><div class="t m0 x11 hb y76 ff7 fs5 fc3 sc0 ls0 ws0">eep learning, rather than a hay understanding of a long laundry</div><div class="t m0 xc hb y77 ff7 fs5 fc3 sc0 ls0 ws0">list of ideas. f youve understood the core ideas well, you can</div><div class="t m0 xc hb y78 ff7 fs5 fc3 sc0 ls0 ws0">rapidly understand other new material. n programming language</div><div class="t m0 xc hb y79 ff7 fs5 fc3 sc0 ls0 ws0">terms, think of it as mastering the core synta, libraries and data</div><div class="t m0 xc hb y7a ff7 fs5 fc3 sc0 ls0 ws0">structures of a new language. ou may still only know a tiny</div><div class="t m0 xc hc y2b ff8 fs6 fc5 sc0 ls0 ws0">at tis ook is aout</div><div class="t m0 xe hd y2c ff7 fs7 fc4 sc0 ls0 ws0">Neural Networks and Deep earning</div><div class="t m0 xf hd y2d ff7 fs7 fc4 sc0 ls0 ws0">hat this book is about</div><div class="t m0 xf hd y2e ff7 fs7 fc4 sc0 ls0 ws0">On the eercises and problems</div><div class="t m0 xf hd y2f ff7 fs7 fc4 sc0 ls0 ws0">sing neural nets to recognie</div><div class="t m0 xf hd y30 ff7 fs7 fc4 sc0 ls0 ws0">handwritten digits</div><div class="t m0 xf hd y7b ff7 fs7 fc4 sc0 ls0 ws0">ow the backpropagation</div><div class="t m0 xf hd y7c ff7 fs7 fc4 sc0 ls0 ws0">algorithm works</div><div class="t m0 xf hd y33 ff7 fs7 fc4 sc0 ls0 ws0">mproving the way neural</div><div class="t m0 xf hd y34 ff7 fs7 fc4 sc0 ls0 ws0">networks learn</div><div class="t m0 xf hd y35 ff7 fs7 fc4 sc0 ls0 ws0"> visual proof that neural nets can</div><div class="t m0 xf hd y36 ff7 fs7 fc4 sc0 ls0 ws0">compute any function</div><div class="t m0 xf hd y37 ff7 fs7 fc4 sc0 ls0 ws0">hy are deep neural networks</div><div class="t m0 xf hd y38 ff7 fs7 fc4 sc0 ls0 ws0">hard to train</div><div class="t m0 xf hd y39 ff7 fs7 fc4 sc0 ls0 ws0">Deep learning</div><div class="t m0 xf hd y3a ff7 fs7 fc4 sc0 ls0 ws0">ppendi: s there a <span class="ff6">siple</span></div><div class="t m0 xf hd y3b ff7 fs7 fc4 sc0 ls0 ws0">algorithm for intelligence</div><div class="t m0 xf hd y3c ff7 fs7 fc4 sc0 ls0 ws0">cknowledgements</div><div class="t m0 xf hd y3d ff7 fs7 fc4 sc0 ls0 ws0">Frequently sked uestions</div><div class="t m0 xe hd y3e ff7 fs7 fc3 sc0 ls0 ws0">f you benefit from the book, please</div><div class="t m0 xe hd y3f ff7 fs7 fc3 sc0 ls0 ws0">make a small donation.  suggest ,</div><div class="t m0 xe hd y40 ff7 fs7 fc3 sc0 ls0 ws0">but you can choose the amount.</div><div class="t m0 xe he y7d ff9 fs8 fc0 sc0 ls0 ws0">Sponsors</div><div class="t m0 xe hd y7e ff7 fs7 fc3 sc0 ls0 ws0">Thanks to all the <span class="fc4">supporters</span> who</div><div class="t m0 xe hd y7f ff7 fs7 fc3 sc0 ls0 ws0">made the book possible, with</div><div class="t m0 xe hd y80 ff7 fs7 fc3 sc0 ls0 ws0">especial thanks to avel Dudrenov.</div><div class="t m0 xe hd y81 ff7 fs7 fc3 sc0 ls0 ws0">Thanks also to all the contributors to</div><div class="t m0 xe hd y82 ff7 fs7 fc3 sc0 ls0 ws0">the <span class="fc4">ugfinder all of Fame</span>.</div><div class="t m0 xe he y83 ff9 fs8 fc0 sc0 ls0 ws0">Resources</div><div class="t m0 xe hd y84 ff7 fs7 fc4 sc0 ls0 ws0">ook F</div><div class="t m0 xe hd y85 ff7 fs7 fc4 sc0 ls0 ws0">Code repository</div><div class="t m0 xe hd y86 ff7 fs7 fc4 sc0 ls0 ws0">ichael Nielsens project</div><div class="t m0 xe hd y87 ff7 fs7 fc4 sc0 ls0 ws0">announcement mailing list</div><div class="t m0 xe hd y88 ff7 fs7 fc4 sc0 ls0 ws0">Deep earning<span class="fc3">, draft book in</span></div><div class="t m0 xe hd y89 ff7 fs7 fc3 sc0 ls0 ws0">preparation, by oshua engio, an</div><div class="t m0 xe hd y8a ff7 fs7 fc3 sc0 ls0 ws0">oodfellow, and aron Courville</div></div><div class="c x10 y4f w5 hf"><div class="t m0 x0 h10 y50 ffa fs9 fc0 sc0 ls0 ws0">欧拉的博客<span class="ffb">:www.liuhao.me</span></div></div></div><div class="pi" data-data='{"ctm":[1.611639,0.000000,0.000000,1.611639,0.000000,0.000000]}'></div></div>
<div id="pf5" class="pf w0 h0" data-page-no="5"><div class="pc pc5 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/62466b0a54457c44d0878d59/bg5.jpg"><div class="c xa y1 w3 h8"><div class="t m0 xb h9 y1a ff5 fs4 fc0 sc0 ls0 ws0">2016/10/10<span class="_ _3"> </span>Neural networks and deep learning</div><div class="t m0 xb h9 y1b ff5 fs4 fc0 sc0 ls0 ws0">http://neuralnetworksanddeeplearning.com/aout.html<span class="_ _4"> </span>2/</div></div><div class="c x5 y1c w4 ha"><div class="t m0 xc hb y8b ff7 fs5 fc3 sc0 ls0 ws0">fraction of the total language - many languages have enormous</div><div class="t m0 xc hb y8c ff7 fs5 fc3 sc0 ls0 ws0">standard libraries - but new libraries and data structures can be</div><div class="t m0 xc hb y8d ff7 fs5 fc3 sc0 ls0 ws0">understood quickly and easily.</div><div class="t m0 xc hb y8e ff7 fs5 fc3 sc0 ls0 ws0">This means the book is emphatically not a tutorial in how to use</div><div class="t m0 xc hb y8f ff7 fs5 fc3 sc0 ls0 ws0">some particular neural network library. f you mostly want to learn</div><div class="t m0 xc hb y90 ff7 fs5 fc3 sc0 ls0 ws0">your way around a library, dont read this book Find the library you</div><div class="t m0 xc hb y91 ff7 fs5 fc3 sc0 ls0 ws0">wish to learn, and work through the tutorial</div><div class="t m0 x14 hb y92 ff7 fs5 fc3 sc0 ls0 ws0">s and documentation.</div><div class="t m0 xc hb y93 ff7 fs5 fc3 sc0 ls0 ws0">ut be warned. hile this has an immediate problem-solving</div><div class="t m0 xc hb y94 ff7 fs5 fc3 sc0 ls0 ws0">payoff, if you want to understand whats really going on in neural</div><div class="t m0 xc hb y95 ff7 fs5 fc3 sc0 ls0 ws0">networks, if you want insights that will still be relevant years from</div><div class="t m0 xc hb y96 ff7 fs5 fc3 sc0 ls0 ws0">now, then its not enough just to learn some hot library. ou need to</div><div class="t m0 xc hb y97 ff7 fs5 fc3 sc0 ls0 ws0">understand the durable, lasting insights underlying how neural</div><div class="t m0 xc hb y98 ff7 fs5 fc3 sc0 ls0 ws0">networks work. Technologies come and technologies go, bu</div><div class="t m0 x15 hb y99 ff7 fs5 fc3 sc0 ls0 ws0">t insight</div><div class="t m0 xc hb y9a ff7 fs5 fc3 sc0 ls0 ws0">is forever.</div><div class="t m0 xc h12 y9b ff8 fsb fc4 sc0 ls0 ws0"> andson approa</div><div class="t m0 xc hb y9c ff7 fs5 fc3 sc0 ls0 ws0">ell learn the core principles behind neural networks and deep</div><div class="t m0 xc hb y9d ff7 fs5 fc3 sc0 ls0 ws0">learning by attacking a concrete problem: the problem of teaching a</div><div class="t m0 xc hb y9e ff7 fs5 fc3 sc0 ls0 ws0">computer to recognie handwritten digits. This problem is</div><div class="t m0 xc hb y9f ff7 fs5 fc3 sc0 ls0 ws0">etremely difficult to solve using the conventional approach to</div><div class="t m0 xc hb ya0 ff7 fs5 fc3 sc0 ls0 ws0">programming. nd yet, as well see, it can be solved pretty well</div><div class="t m0 xc hb ya1 ff7 fs5 fc3 sc0 ls0 ws0">using a simple neural network, with just a few </div><div class="t m0 x16 hb ya2 ff7 fs5 fc3 sc0 ls0 ws0">tens of lines of code,</div><div class="t m0 xc hb ya3 ff7 fs5 fc3 sc0 ls0 ws0">and no special libraries. hats more, well improve the program</div><div class="t m0 xc hb ya4 ff7 fs5 fc3 sc0 ls0 ws0">through many iterations, gradually incorporating more and more of</div><div class="t m0 xc hb ya5 ff7 fs5 fc3 sc0 ls0 ws0">the core ideas about neural networks and deep learning.</div><div class="t m0 xc hb ya6 ff7 fs5 fc3 sc0 ls0 ws0">This hands-on approach means that youll need some programming</div><div class="t m0 xc hb ya7 ff7 fs5 fc3 sc0 ls0 ws0">eperience to read the book. ut you dont need to be a professional</div><div class="t m0 xc hb ya8 ff7 fs5 fc3 sc0 ls0 ws0">programmer. ve written the code in ython version ., which</div><div class="t m0 x17 hb ya9 ff7 fs5 fc3 sc0 ls0 ws0">,</div><div class="t m0 xc hb yaa ff7 fs5 fc3 sc0 ls0 ws0">even if you dont program in ython, should be easy to understand</div><div class="t m0 xc hb yab ff7 fs5 fc3 sc0 ls0 ws0">with just a little effort. Through the course of the book we will</div><div class="t m0 xc hb yac ff7 fs5 fc3 sc0 ls0 ws0">develop a little neural network library, which you can use to</div><div class="t m0 xc hb yad ff7 fs5 fc3 sc0 ls0 ws0">eperiment and to build understanding. ll the code is available for</div><div class="t m0 xc hb yae ff7 fs5 fc3 sc0 ls0 ws0">download <span class="fc4">here</span>. Once youve finished the book, or as you read it, you</div><div class="t m0 xc hb yaf ff7 fs5 fc3 sc0 ls0 ws0">can easily pick up one of the more feature-complete neural network</div><div class="t m0 xc hb yb0 ff7 fs5 fc3 sc0 ls0 ws0">librar</div><div class="t m0 x18 hb yb1 ff7 fs5 fc3 sc0 ls0 ws0">ies intended for use in production.</div><div class="t m0 xc hb yb2 ff7 fs5 fc3 sc0 ls0 ws0">On a related note, the mathematical requirements to read the book</div><div class="t m0 xc hb yb3 ff7 fs5 fc3 sc0 ls0 ws0">are modest. There is some mathematics in most chapters, but its</div><div class="t m0 xe hd yb4 ff7 fs7 fc3 sc0 ls0 ws0">y <span class="fc4">ichael Nielsen</span>  an 1</div></div><div class="c x10 y4f w5 hf"><div class="t m0 x0 h10 y50 ffa fs9 fc0 sc0 ls0 ws0">欧拉的博客<span class="ffb">:www.liuhao.me</span></div></div></div><div class="pi" data-data='{"ctm":[1.611639,0.000000,0.000000,1.611639,0.000000,0.000000]}'></div></div>
神经网络与深度学习
