前言 Preface

Deep learning is driving the AI revolution and PyTorch is making it easier than ever before for anyone to build deep learning applications.

深度学习正在促使 AI 革命, 并且 PyTorch 使得它相较以往构建深度学习应用变得更容易.

语法结构 and 连接两个简单句.
Deep learning is driving the AI revolution.
主谓宾 (S+V+O) 结构, 现在进行时, 主动语态.
句子成分语法分析
主语: Deep learning 名词短语.
谓语: is driving 动词短语, 现在进行时. 助动词+实义动词, drive 为 vt., 后接宾语. 译为: 驱动, 推动. drive 科技类文章高频词, 替代普通动词 promote: 促进/推进.
宾语: the AI revolution 名词短语. revolution 革新/革命. 搭配: technological revolution (技术革新), industrial revolution (工业革命)
扩展 - 举一反三:
Large language models are driving the development of natural language processing.
Computer vision is driving the innovation of autonomous driving.
Cloud computing is driving the digital transformation of enterprises.
扩展 - 同义改写:
fuel, v 加燃料, 加油, 刺激. 及物动词, 可替代 drive.
Deep learning is fueling the AI revolution.
push forward, 动词短语, 推进, 强调动作的持续性.
Deep learning is pushing forward the AI revolution.
名词化改写, 更加正式: driving force behind sth., 什么的背后推动力, force 这里指力量, driving 修饰力量, 表推动的力量.
Deep learning is the driving force behind the AI revolution.
PyTorch is making it easier than ever before for anyone to build deep learning applications.
主谓 + 形式宾语 + 真正宾语 (for sb. to do). 现在进行时, 主动语态.
句子成分语法分析
主语: PyTorch 专有名词
谓语: is making 动词短语, 助动词 + 实义动词, make 使得...,
形式宾语: it 代替后面的真正宾语, 平衡句子, 避免 making 后跟过长宾语.
宾语补足语: easer than ever before 比较级 + 比较状语: than ever before, 比以往都容易 (easer).
真正的宾语: for anyone to build deep learning applications 不定式复合结构 (for sb. + to do). it 指代的内容.
表范围的状语: for anyone 介词短语, 限定动作对象, 任何人.
表比较的状语: than ever before 比较状语短语, 强调比较级的程度: 比以往任何时候.
句子的难点是 形式宾语 it 的用法. 平衡句子结构. 句型结构: make it + adj. + for sb. + to do sth.
make 使役动词, 使得, 让. 后接复合宾语 (宾语 + 宾语补足语).
it 形式宾语, 必须使用代词 it. 仅语法作用, 无实际意义.
adj. 形容词(原级/比较级), 作宾补, 说明做事的状态.
for sb. 逻辑主语, 说明谁来做. 若 sb. 是动作的发出者, 必需使用 for, 不用 of.
to do sth. 真正的宾语. 句子的核心表意部分.
例句:
she makes it easy for me to learn English.
扩展 - 举一反三:
TensorFlow is making it faster than ever before for engineers to deploy AI models.
Python is making it more accessible for beginners to enter the field of data science.
Cloud GPUs are making it cheaper than ever before for small teams to train large models.
扩展 - 同义改写:
render 正式使用, 表使得, 使成为. 代替 make.
PyTorch is rendering it easier than ever for anyone to develop deep learning applications.
改写为现在完成时: have + done, 强调已实现的效果.
PyTorch has simplified the process of building deep learning applications for anyone, more so than ever before.
调整语序, 突出结果.
Building deep learning applications has become easier than ever for anyone, thanks to PyTorch.
easier than ever before 是强调程度的比较级, 科技文中常用.

句子成分	语法分析
主语: Deep learning	名词短语.
谓语: is driving	动词短语, 现在进行时. 助动词+实义动词, drive 为 vt., 后接宾语. 译为: 驱动, 推动. drive 科技类文章高频词, 替代普通动词 promote: 促进/推进.
宾语: the AI revolution	名词短语. revolution 革新/革命. 搭配: technological revolution (技术革新), industrial revolution (工业革命)

句子成分	语法分析
主语: PyTorch	专有名词
谓语: is making	动词短语, 助动词 + 实义动词, make 使得...,
形式宾语: it	代替后面的真正宾语, 平衡句子, 避免 making 后跟过长宾语.
宾语补足语: easer than ever before	比较级 + 比较状语: than ever before, 比以往都容易 (easer).
真正的宾语: for anyone to build deep learning applications	不定式复合结构 (for sb. + to do). it 指代的内容.
表范围的状语: for anyone	介词短语, 限定动作对象, 任何人.
表比较的状语: than ever before	比较状语短语, 强调比较级的程度: 比以往任何时候.

This book will help you uncover expert techniques and gain insights to get the most out of your data and build complex neural network models.

本书会帮助你通过探索专业技术, 并获悉洞察力, 来尽可能地利用数据, 并构建复杂的神经网络模型.

并列简单句. 含多层并列结构 + 目的状语. 主要语法结构: 主谓宾 + 宾补 (并列不定式) + 目的状语 (并列不定式). 一般将来时, 主动语态.
句子成分语法分析
主语: this book 名词短语.
谓语: will help 一般将来时.
宾语: you 人称代词, 宾格.
宾补 (并列): uncover expert techniques and gain insights 并列不定式结构. help sb. (to) do sth. 其中 to 可以省略. 表示帮助某某做什么. uncover/gain 并列动词, 分别接各自宾语 expert techniques/insights.
目的状语 (并列): to get the most out of your data and build complex neural network models 并列不定式结构 (to + v. and v.), 并列动词作 uncover/gain 的目的状语.
核心语法: help sb. (to) do sth.. 可省略 to, 也可以不省略, 句子含义一样. 例句: This tool helps (to) speed up model training.
多层并列结构是本句的难点.
宾补并列. 两个动作并列, 作 help you 的宾补. 其中 uncover 可翻译为发现/探索. gain insights 可译为获得洞悉/深入理解. gain insights 后可跟 into sth. 例如: gain insights into NN (深入理解神经网络).
目的状语并列. 一个 to (共用) + 两个动词并列, 作前面两个动词的目的状语. 这是不定式做状语, 常用句式为 "动作 + 目的", 为了强调目的, 也可将 to 改写为 in order to
This book will help you uncover expert techniques and gain insights in order to get the most out of your data and build complex neural network models.
固定用法:
get most out of sth. 充分利用某事物. 类似的有: make full use of.
扩展 - 举一反三. 句型: 主语 will help sb + 并列动词 + to + 并列目的动作. 可以译为: xxx 可以帮助 xxx 做什么, 从而达到 xxx 目的.
This course will help you master Python skills and acquire experience to process big data and develop machine learning models.
This tool will help you automate data cleaning and save time to focus on model optimization and improve prediction accuracy.
This guide will help you understand AI fundamentals and learn practices to solve real-world problems and deploy AI applications.
扩展 - 同义替换.
可使用 enable 代替 help, 行文更正式.

句子成分	语法分析
主语: this book	名词短语.
谓语: will help	一般将来时.
宾语: you	人称代词, 宾格.
宾补 (并列): uncover expert techniques and gain insights	并列不定式结构. help sb. (to) do sth. 其中 to 可以省略. 表示帮助某某做什么. uncover/gain 并列动词, 分别接各自宾语 expert techniques/insights.
目的状语 (并列): to get the most out of your data and build complex neural network models	并列不定式结构 (to + v. and v.), 并列动词作 uncover/gain 的目的状语.

The book starts with a quick overview of PyTorch and explores convolutional neural network (CNN) architectures for image classification.

本书首先会快速介绍 PyTorch, 并探究了适用于图片分类的卷积神经网络 (CNN) 架构.

同意主语, 并列谓语的简单句.
句子中 for 译为: 适用于. for image classification, 适用于图像分类的.
扩展 - 举一反三:
The tutorial starts with a basic introduction to TensorFlow and explores recurrent neural network (RNN) architectures for text generation.
The paper starts with a detailed analysis of transformer models and explores improved architectures for machine translation.

Similarly, you will explore recurrent neural network (RNN) architectures as well as Transformers and use them for sentiment analysis.

类似的, 你还会讨论循环神经网络架构, 以及 Transformers 模型, 并使用它来完成情感分析.

并列谓语简单句. 其中 and use 省略了 will, 原为: and will use ...
as well as, 和/以及, 含义与 and 一样. 使用更加正式, 更强调补充性.
并列谓语部分:
第一个谓语为 explore, 探索, 其宾语为 RNN 和 Transformers
第二个谓语为 use, 使用. 后跟 for sentiment analysis 作状语部分, 用于说明使用场景.
扩展 - 举一反三: 句型 will explore + (will) use. 表示先探索 xxx, 然后使用.
Similarly, you will explore attention mechanisms as well as BERT models and use them for text classification.
Similarly, we will explore LSTM architectures as well as GPT models and use them for text generation.

Next, you will learn how to create arbitrary neural network architectures and build Graph neural networks (GNNs).

然后, 你会学习如何创建自定义神经网络架构, 并构建图神经网络(GNN).

又是并列谓语简单句, will learn ... (will) build... 共用情态动词 will.
此处 arbitrary 不翻译成任意的, 译为自定义的更贴切. 算是 DL 中固定的译法.
扩展 - 举一反三:
Next, you will learn how to design arbitrary model structures and build Graph Convolutional Networks (GCNs).
Next, we will learn how to tune arbitrary hyperparameters and build Graph Attention Networks (GATs).
Next, you will learn how to construct arbitrary data pipelines and build GraphSAGE models for node classification.

As you advance, you’ll apply deep learning (DL) across different domains such as music, text, and image generation using generative models including Generative adversarial networks (GANs) and diffusion.

随着学习的逐步深入, 你会将 DL 应用于不同领域, 诸如音乐, 文本, 以及图像生成等. 会使用到生成式模型, 其中包括 GANs 与扩散模型.

时间状语 + 主句结构, 主语嵌套伴随方式状语.
句子成分说明
时间状语从句: as you advance 其中 as 表示随着, 衔接进阶学习 (advance, v. 进阶).
主句: you'll apply DL across different domain such as music, text, and image generation. 主语: you. 谓语: will apply, 将会应用. 宾语: DL. 范围状语: across different domain, 表示应用范围. 后置定语: such as ..., 限定/修饰 domain.
伴随方式状语: using generative models ... using 现在分词引导, including ... 为后置定语, 修饰 models. 其中 GANs 为固定写法, 需要加 s.
句型: apply ... across different domain, 跨领域应用.
扩展 - 举一反三:
As you advance, you’ll apply machine learning (ML) across different domains such as fraud detection, recommendation systems, and speech recognition using supervised models including logistic regression and support vector machines (SVMs).
As you advance, you’ll apply Graph Neural Networks (GNNs) across different domains such as social network analysis, knowledge graph completion, and drug discovery using graph generative models including GraphGAN and diffusion GNNs.
As you advance, you’ll apply large language models (LLMs) across different domains such as machine translation, chatbots, and content creation using pre-trained models including GPT and BERT.

句子成分	说明
时间状语从句: as you advance	其中 as 表示随着, 衔接进阶学习 (advance, v. 进阶).
主句: you'll apply DL across different domain such as music, text, and image generation.	主语: you. 谓语: will apply, 将会应用. 宾语: DL. 范围状语: across different domain, 表示应用范围. 后置定语: such as ..., 限定/修饰 domain.
伴随方式状语: using generative models ...	using 现在分词引导, including ... 为后置定语, 修饰 models. 其中 GANs 为固定写法, 需要加 s.

补充说明:
状语是句子的修饰成分. 其核心作用是给句子的核心动作 (动词), 状态 (表语) 或整个句子补充信息, 包括: 怎么来的, 什么时候, 什么地方, 到什么程度, 处于什么原因, 以及目的等信息.
谓语表示动作, 是要做什么. 状语就是在说怎么做, 何时做, 何地做, 为何做.
状语修饰的是动词, 形容词, 副词, 以及整个句子. 逻辑上删除后对句子主干不影响.
判断状语的依据
修饰性.
可删性.
技术类文章常见的 6 种状语
方式状语. 表示以什么方式/手段 (补充动词, 说明怎么干). 常用词汇: using/by/with
You'll build GNNs using custom architectures.
主干: 你将构建图神经网络, 其中 using 引导伴随状语, 修饰 build, 补充说明, 采用自定义架构的方式.
范围/地点状语. 表示在什么范围/场地/领域. 用于说明在哪干. 常用词汇: across/in/on/at
We apply DL in image classification.
in 引导范围状语, 用于说明 apply 的应用领域.
时间状语. 说明在什么时候干. 常用词汇: next/then/as/when/after/before 等.
Next, you'll learn to create NN architecture.
目的状语. 表示动作是为了什么, 为啥干. 常用词汇: for/to (do sth to do sth/use sth for sth)
Web use CNNs for image classification.
程度状语. 常用词汇: very/quite/highly/fully.
GANs are widely used in image generation.
结果状语. 表示干了之后会怎样. 常用词汇: thus/therefore/so, doing (现在分词引导)
We train the model for 200 epochs, achieving 99% accuracy.
for 引导程度状语修饰 train. acheving 现在分词引导结果状语, 补充说明 train 的结果.
特殊状语: 伴随状语. 用于修饰主句中核心动作/主语状态, 表示在主句动作发生的同时, 同步存在另一个动作/状态/方式. 目的是补充主句进而说明以什么方式/在什么状态下/伴随什么动作发生.
核心特点: 依附主句存在, 无独立主谓结构. 删除后句子主干与核心意思不变. 多半以现在分词引导.
句子核心: You'll apply DL using generative models. 主句动词是应用 apply, using 表示以使用生成式模型的方式来应用.
核心判断特征:
同时性, 与主句动作同时存在, 无先后.
依附性, 没有独立主谓结构, 必须跟着主句走.
修饰性, 只对主句或主句的动作进行补充.
可删性. 删除后对主句的结构与含义不影响.
技术类文章中常见的用法
现在分词作伴随状语. 表主动. 伴随状语的动作发起者(逻辑主语)与主句主语一致.
We train the model for 100 epochs, obtaining a test accuracy of 98%. (基于预训练大语言模型，我们针对文本分类微调该模型。)
介词短语作伴随状语. 表依据/基础/条件.
Based on the pre-trained LLM, we fine-tune the model for text classification.
We analyze the experimental results according to the loss curve and confusion matrix. (我们根据损失曲线和混淆矩阵分析实验结果。)
过去分词作伴随状语, 表示被动.
The dataset is split into train and test sets, processed by data augmentation and normalization. (该数据集被划分为训练集和测试集，同时经过了数据增强和归一化处理。)

Next, you’ll build and train your own deep reinforcement learning models in PyTorch, as well as interpreting DL models.

下一步, 你将使用 PyTorch 来构建并训练你的深度强化学习模型, 同时学习解释 DL 模型.

as well as, 表示同时, 和的意思. 这里 interpreting 翻译为解释, 主要是因为有一个可解释性模型 (XAI) 的专业方向. 需要注意的是, 它不同于 and, and 表示并列句, 这里逻辑上是 and 的含义, 但是由于 as well as 是介词短语, 后面跟动名词.
翻译时不直接翻译为 "解释", 而是翻译成 "学习解释". 逻辑上更贴切.
该句是典型的教程类文章中的句式, "核心动作 + 补充动作", 逻辑上是先掌握核心内容, 在补充配套技能.
扩展 - 举一反三
Next, you’ll build and train your own reinforcement learning agents in TensorFlow, as well as evaluating model performance.
Next, we’ll build and train your own DQN models in PyTorch, as well as visualizing the reward function (奖励函数).

You will not only learn how to build models but also how to deploy them into production and to mobile devices (Android and iOS) using expert tips and techniques.

你不仅仅会学习到如何构建模型, 还会学习到如何将其部署到生产环境, 以及移动设备中 (安卓和 iOS), 在此期间将会使用到业内资深的实操技巧与方法.

这里 expert tips and techniques 翻译为业内资深的实操技巧与方法.
using 引导伴随状语, 修饰 deploy, 表方式.
not only... but also, 递进式语法结构.
and to mobile ... 与前置的 to deploy them into production 并列, 省略了 to deploy them 部分.
扩展 - 举一反三
You will not only learn how to train models but also how to optimize them for low-latency inference and to edge devices (Raspberry Pi and Jetson) using industry best practices. (你不仅会学习如何训练模型，还会掌握如何借助行业最佳实践，将这些模型优化以实现低延迟推理，并部署到树莓派、Jetson 等边缘设备中。)
You will not only learn how to design LLMs but also how to deploy them into cloud services and to embedded systems using optimization techniques. (你不仅会学习如何设计大语言模型，还会掌握如何借助优化方法，将这些模型部署到云服务以及嵌入式系统中。)

Next, you will master the skills of training large models efficiently in a distributed fashion, searching neural architectures effectively with AutoML, as well as rapidly prototyping models using fastai.

然后, 你将会精通三个技能, 包括基于分布式的大模型训练, 使用 AutoML 来完成神经搜索架构, 以及通过 fastai 框架快速地进行模型的原型开发.

句子采用长宾语结构, 翻译时对句子进行了分解, 避免直译.
句子主干: you will master the skills of ... 这里 master 作动词, 表示精通. 句子 of 后为精通的内容, 作宾语.
三个动词的现在分词做宾语, 表主动:
句子成分说明
training large models efficiently in a distributed fashion
searching neural architectures effectively with AutoML
repidly prototyping models use fastai
training, 训练.
searching, 搜索.
prototyping, prototype 作动词, 制作原型.

句子成分	说明
training large models efficiently in a distributed fashion
searching neural architectures effectively with AutoML
repidly prototyping models use fastai