深度学习入门基于Python的理论与实现

Name: 深度学习入门 基于Python的理论与实现
Brand: 人民邮电出版社
SKU: X95-29248305
Price: 46.87 USD
Availability: InStock

Brand: 人民邮电出版社 Sold By: EM China

$46.87 In Stock

Ships from China Arrives in 10-15 business days

Item can be returned in its original condition for a full refund or replacement within 30 days of receipt.

Delivery:

July2026

Delivery Guarantee

$5 Credit If Delayed

Full Refund If Item Is Damaged

15-Day Full Refund If No Updates

30-Day Full Refund If Not Delivered

Product Details

产品特色

编辑推荐

1.日本深度学习入门经典畅销书，原版上市不足2年印刷已达100 000册。长期位列日亚“人工智能”类图书榜首，众多五星好评。

2.使用Python 3，尽量不依赖外部库或工具，从零创建一个深度学习模型。

3.示例代码清晰，源代码可下载，需要的运行环境非常简单。读者可以一边读书一边执行程序，简单易上手。

4.使用平实的语言，结合直观的插图和具体的例子，将深度学习的原理掰开揉碎讲解，简明易懂。

5.使用计算图介绍复杂的误差反向传播法，非常直观。

6.相比AI圣经“花书”，本书更合适入门。

对于非AI方向的技术人员，本书将大大降低入门深度学习的门槛；对于在校大学生、研究生，本书不失为学习深度学习的一本好教材；即便是在工作中已经熟练使用框架开发各类深度学习模型的读者，也可以从本书中获得新的体会。——摘自本书译者序

内容简介

本书是深度学习真正意义上的入门书，深入浅出地剖析了深度学习的原理和相关技术。书中使用Python3，尽量不依赖外部库或工具，从基本的数学知识出发，带领读者从零创建一个经典的深度学习网络，使读者在此过程中逐步理解深度学习。书中不仅介绍了深度学习和神经网络的概念、特征等基础知识，对误差反向传播法、卷积神经网络等也有深入讲解，此外还介绍了深度学习相关的实用技巧，自动驾驶、图像生成、强化学习等方面的应用，以及为什么加深层可以提高识别精度等疑难的问题。

本书适合深度学习初学者阅读，也可作为高校教材使用。

作者简介

斋藤康毅（作者）东京工业大学毕业，并完成东京大学研究生院课程。现从事计算机视觉与机器学习相关的研究和开发工作。是Introducing Python、Python in Practice、The Elements of Computing Systems、Building Machine Learning Systems with Python的日文版译者。陆宇杰（译者）众安科技NLP算法工程师。主要研究方向为自然语言处理及其应用，对图像识别、机器学习、深度学习等领域有密切关注。Python爱好者。

目　　录

目录

译者序　xiii

前言　xv

第　1 章 Python入门　1

1.1　Python是什么　1

1.2　Python的安装　2

1.2.1　Python版本　2

目录 译者序　xiii 前言　xv 第　1 章 Python入门　1 1.1　Python是什么　1 1.2　Python的安装　2 1.2.1　Python版本　2 1.2.2　使用的外部库　2 1.2.3　Anaconda发行版　3 1.3　Python解释器　4 1.3.1　算术计算　4 1.3.2　数据类型　5 1.3.3　变量　5 1.3.4　列表　6 1.3.5　字典　7 1.3.6　布尔型　7 1.3.7　if 语句　8 1.3.8　for 语句　8 1.3.9　函数　9 1.4　Python脚本文件　9 1.4.1　保存为文件　9 1.4.2　类　10 1.5　NumPy　11 1.5.1　导入NumPy　11 1.5.2　生成NumPy数组　12 1.5.3　NumPy 的算术运算　12 1.5.4　NumPy的N维数组　13 1.5.5　广播　14 1.5.6　访问元素　15 1.6　Matplotlib　16 1.6.1　绘制简单图形　16 1.6.2　pyplot 的功能　17 1.6.3　显示图像　18 1.7　小结　19 第　2 章 感知机　21 2.1　感知机是什么　21 2.2　简单逻辑电路　23 2.2.1　与门　23 2.2.2　与非门和或门　23 2.3　感知机的实现　25 2.3.1　简单的实现　25 2.3.2　导入权重和偏置　26 2.3.3　使用权重和偏置的实现　26 2.4　感知机的局限性　28 2.4.1　异或门　28 2.4.2　线性和非线性　30 2.5　多层感知机　31 2.5.1　已有门电路的组合　31 2.5.2　异或门的实现　33 2.6　从与非门到计算机　35 2.7　小结　36 第3　章 神经网络　37 3.1　从感知机到神经网络　37 3.1.1　神经网络的例子　37 3.1.2　复习感知机　38 3.1.3　激活函数登场　40 3.2　激活函数　42 3.2.1　sigmoid 函数　42 3.2.2　阶跃函数的实现　43 3.2.3　阶跃函数的图形　44 3.2.4　sigmoid 函数的实现　45 3.2.5　sigmoid 函数和阶跃函数的比较　46 3.2.6　非线性函数　48 3.2.7　ReLU函数　49 3.3　多维数组的运算　50 3.3.1　多维数组　50 3.3.2　矩阵乘法　51 3.3.3　神经网络的内积　55 3.4　3 层神经网络的实现　56 3.4.1　符号确认　57 3.4.2　各层间信号传递的实现　58 3.4.3　代码实现小结　62 3.5　输出层的设计　63 3.5.1　恒等函数和softmax 函数　64 3.5.2　实现softmax 函数时的注意事项　66 3.5.3　softmax 函数的特征　67 3.5.4　输出层的神经元数量　68 3.6　手写数字识别　69 3.6.1　MNIST数据集　70 3.6.2　神经网络的推理处理　73 3.6.3　批处理　75 3.7　小结　79 第4　章 神经网络的学习　81 4.1　从数据中学习　81 4.1.1　数据驱动　82 4.1.2　训练数据和测试数据　84 4.2　损失函数　85 4.2.1　均方误差　85 4.2.2　交叉熵误差　87 4.2.3　mini-batch 学习　88 4.2.4　mini-batch 版交叉熵误差的实现　91 4.2.5　为何要设定损失函数　92 4.3　数值微分　94 4.3.1　导数　94 4.3.2　数值微分的例子　96 4.3.3　偏导数　98 4.4　梯度　100 4.4.1　梯度法　102 4.4.2　神经网络的梯度　106 4.5　学习算法的实现　109 4.5.1　2 层神经网络的类　110 4.5.2　mini-batch 的实现　114 4.5.3　基于测试数据的评价　116 4.6　小结　118 第5　章 误差反向传播法　121 5.1　计算图　121 5.1.1　用计算图求解　122 5.1.2　局部计算　124 5.1.3　为何用计算图解题　125 5.2　链式法则　126 5.2.1　计算图的反向传播　127 5.2.2　什么是链式法则　127 5.2.3　链式法则和计算图　129 5.3　反向传播　130 5.3.1　加法节点的反向传播　130 5.3.2　乘法节点的反向传播　132 5.3.3　苹果的例子　133 5.4　简单层的实现　135 5.4.1　乘法层的实现　135 5.4.2　加法层的实现　137 5.5　激活函数层的实现　139 5.5.1　ReLU层　139 5.5.2　Sigmoid 层　141 5.6　Affine/Softmax层的实现　144 5.6.1　Affine层　144 5.6.2　批版本的Affine层　148 5.6.3　Softmax-with-Loss 层　150 5.7　误差反向传播法的实现　154 5.7.1　神经网络学习的全貌图　154 5.7.2　对应误差反向传播法的神经网络的实现　155 5.7.3　误差反向传播法的梯度确认　158 5.7.4　使用误差反向传播法的学习　159 5.8　小结　161 第6　章 与学习相关的技巧　163 6.1　参数的更新　163 6.1.1　探险家的故事　164 6.1.2　SGD　164 6.1.3　SGD的缺点　166 6.1.4　Momentum　168 6.1.5　AdaGrad　170 6.1.6　Adam　172 6.1.7　使用哪种更新方法呢　174 6.1.8　基于MNIST数据集的更新方法的比较　175 6.2　权重的初始值　176 6.2.1　可以将权重初始值设为0 吗　176 6.2.2　隐藏层的激活值的分布　177 6.2.3　ReLU的权重初始值　181 6.2.4　基于MNIST数据集的权重初始值的比较　183 6.3　Batch Normalization　184 6.3.1　Batch Normalization 的算法　184 6.3.2　Batch Normalization 的评估　186 6.4　正则化　188 6.4.1　过拟合　189 6.4.2　权值衰减　191 6.4.3　Dropout　192 6.5　超参数的验证　195 6.5.1　验证数据　195 6.5.2　超参数的化　196 6.5.3　超参数化的实现　198 6.6　小结　200 第7　章 卷积神经网络　201 7.1　整体结构　201 7.2　卷积层　202 7.2.1　全连接层存在的问题　203 7.2.2　卷积运算　203 7.2.3　填充　206 7.2.4　步幅　207 7.2.5　3 维数据的卷积运算　209 7.2.6　结合方块思考　211 7.2.7　批处理　213 7.3　池化层　214 7.4　卷积层和池化层的实现　216 7.4.1　4 维数组　216 7.4.2　基于im2col 的展开　217 7.4.3　卷积层的实现　219 7.4.4　池化层的实现　222 7.5　CNN的实现　224 7.6　CNN的可视化　228 7.6.1　第 1 层权重的可视化　228 7.6.2　基于分层结构的信息提取　230 7.7　具有代表性的CNN　231 7.7.1　LeNet　231 7.7.2　AlexNet　232 7.8　小结　233 第8　章 深度学习　235 8.1　加深网络　235 8.1.1　向更深的网络出发　235 8.1.2　进一步提高识别精度　238 8.1.3　加深层的动机　240 8.2　深度学习的小历史　242 8.2.1　ImageNet　243 8.2.2　VGG　244 8.2.3　GoogLeNet　245 8.2.4　ResNet　246 8.3　深度学习的高速化　248 8.3.1　需要努力解决的问题　248 8.3.2　基于GPU的高速化　249 8.3.3　分布式学习　250 8.3.4　运算精度的位数缩减　252 8.4　深度学习的应用案例　253 8.4.1　物体检测　253 8.4.2　图像分割　255 8.4.3　图像标题的生成　256 8.5　深度学习的未来　258 8.5.1　图像风格变换　258 8.5.2　图像的生成　259 8.5.3　自动驾驶　261 8.5.4　Deep Q-Network（强化学习）　262 8.6　小结　264 附录A　Softmax-with-Loss 层的计算图　267 A.1　正向传播　268 A.2　反向传播　270 A.3　小结　277 参考文献　279

显示全部信息

Shipping & Delivery

Shipping Overview:

• Shipping: Standard Domestic Shipping within the United States charges first item USD 6.99 and additional item USD 3.0 each. Standard International Shipping from United Kingdom, Germany, Italian and Japan to the United States charges first item USD 11.99 and additional item USD 4.0 each.

• Import Duty: Everymarket collects applicable duties and taxes on goods imported into the U.S. at checkout, it's calculated based on the goods value multiply by the rate. For United Kingdom and China goods is 10%, Germany, Italian and Japan goods is 15%. We are committed to making your U.S. shopping experience simple and transparent. The method of delivery—either Delivered Duty Paid (DDP) or Delivered Duty Unpaid (DDU)—depends on whether duties and taxes are collected at the time of your purchase.

• Order Processing: Please allow 1-2 business days for order processing and preparation before shipment.

• Domestic Shipping: Orders within the U.S. are shipped via USPS or FedEx, depending on the origin of the product. Average transit times to the United States is 3 - 7 Business days and average transit times to International is 7 - 14 Business days

• International Shipping: Currently, we only ship within the USA.

• Tracking Information: Every order is trackable. You will receive a tracking number once your order has been shipped. Products may be shipped from various global fulfillment centers.

Shipping Delays:

Please note that shipping times may vary due to factors beyond our control, such as weather conditions, natural disasters, or peak holiday periods. While we strive to ensure timely delivery, the exact arrival time cannot be guaranteed and is managed by the shipping carrier.

Shipping Options:

Standard Delivery: Most orders are shipped within 3-7 business days. Larger items may utilize LTL shipping for safe handling.
Handling Time: We handle shipments on business days (Monday - Friday), with a preparation time of 1-2 days.
Additional Charges: Some items require additional shipping charges due to their size, weight, or special handling. These charges are specified on the product pages and are not eligible for shipping discounts.
Exclusions: Gift cards, packaging, taxes, and prior purchases do not count toward the minimum purchase requirement for free shipping. This offer is valid only for shipments to U.S. addresses, including Puerto Rico.

Delivery Details:

Estimates: Standard shipping within the US typically takes 3-7 business days. These are estimates and not guarantees.
Shipping Restrictions: We ship to all 50 states, Washington, DC, U.S. territories, and APO/FPO/DPO addresses. Shipping options vary based on the delivery address.
Remote Areas: Shipments to remote areas may incur additional charges or require pickup from a nearby shipping partner’s location.

Shipping Confirmation:

You will receive a shipping confirmation email with a tracking number as soon as your order is dispatched. If you do not receive this email immediately, please be assured that your items will arrive within the estimated delivery window provided at checkout.

Order Modifications:

If you need to cancel or modify your order, please contact our customer support immediately.

Issues with Delivery:

If your order shows as delivered but you have not received it, please contact the shipping carrier directly to resolve the issue. For persistent problems, contact our customer service at cs@everymarket.com.

Customer Support:

Our team is available 24/7 to assist you with any questions or concerns regarding your order. We are committed to ensuring a smooth shopping experience.

Returns & Refunds

Return & Refund Policy Overview

Please review our return and refund policies below to ensure a smooth transaction process.

Return Policy

Duration: You have 30 days from receiving your item to initiate a return.
Condition: Items must be unworn, unwashed, with original tags and packaging intact.
Shipping Costs: Please contact us to initiate your return, we'll offer return label for free. Otherwise we are not responsible for the return shipping costs of the items.
Packaging: Ensure returned items are well-packaged to avoid damage during transit.
Tracking: Use a trackable and insured shipping method as we are not liable for items lost or damaged on return.
Initiating a Return: Contact us at cs@everymarket.com to start your return. We will provide a return shipping label and instructions upon approval. Returns without prior approval will not be accepted.

How to Return

Method: Returns must be sent back by mail to the address provided in the return instructions.

Return Label

Please contact us to get return labels.

Product Conditions for Return

Eligible Products: Returns are only accepted for items in new condition.

Nonreturnable Items

Certain items are not eligible for return:

Electronic devices after 30 days (e.g., computers, laptops, Kindles)
Gift cards, prepaid game cards
Perishable goods, live insects, some jewelry, some health and personal care items
Customized or personalized products
Items with special shipping restrictions

Refund Policy

Window: Eligible products may be returned within 30 days of delivery for a refund.
Refund Method: Refunds are processed to the original payment method or as store credit for items purchased with gift cards.
Processing Time: Refunds are processed within 3-5 business days after we receive the return; please allow additional time for shipping and bank processing.

Claims

Inspect your order upon arrival and report any defects, damages, or incorrect items immediately to allow us to address the issue. For claims, contact our support team with details of the issue.

Exchange Policy

For the quickest service, return your original item and place a new order for the desired product once your return is accepted.

Return Address

EveryMarket Customer Service 2101 E Terra Ln, O'Fallon, MO 63366

Customer Support

Available 24/7 for any questions or assistance needed:

Phone: +1 (800) 660-3783
Email: cs@everymarket.com