Knowledge Of AI

Knowledge Of AI

目录

Fathers of AI

Development history 

Origin        

First peak

First trough

Second peak

Second trough

Rise again

Last three years

Some key technologies

Turing test

SVM（Support Vector Machine）

Two advantages of SVM：

Specific explanation

Boosting

Deep Learning

Language

Python

Advantages of Python

C++

Lisp

Prolog

Java

Example

metaverse（元宇宙）

Others

Virtual personal assistant

Driverless

online Service

Music and movie recommendation service

Smart home device

---

Luck_ZY will AK IOI

Bright_XZJ has AFO 

Our English teacher asked me to do a popular science on AI.So I browsed a lot of materials, wrote down what I learned, and sorted it out to make the content more organized.

My English is not particularly professional, so I use Baidu translation a lot, so there may be some mistakes.

Fathers of AI

Because of the achievement for AI, there are 4 people who has been named"fathers of AI"

Alan Turing : Founder of computer logic

John McCarthy:Raise the concept of "artificial intelligence"

Marvin Minsky：Together with his classmate Dunn Edmund, he built the world's first neural network computer.

Seymour Paput：Integrating science and technology with education

Development history 

The history of AI has ups and downs.

Origin        

Artificial intelligence was formally proposed in the 1950s and 1960s.

In 1950, a senior named Marvin Minsky (later known as the "father of artificial intelligence") and his classmate Dunn Edmund built the world's first neural network computer as the beginning of artificial intelligence.

At the same time,Alan Turing proposed a method called "Turing test" to test whether a computer is intelligent. Moreover, he boldly predicted the feasibility of the existence of intelligent machines.

 In 1956,computer expert John McCarthy put forward the term "artificial intelligence", which is regarded as the symbol of the official birth of artificial intelligence.

First peak

Computers are widely used in the fields of mathematics and natural language to solve algebra（代数）, geometry（几何） and English problems.

Within twenty years, machines will be able to accomplish everything that people can do.

First trough

Due to the serious lack of understanding of the difficulty of AI development, the cooperation plan with the Defense Advanced Research Program of the United States （美国国防高级研究计划署）failed.

Three technical bottlenecks：

1.Insufficient computer performance

2.Complexity of the problem

3.The amount of data is seriously missing

AI technology has been silent for six years.

Second peak

In 1980, Carnegie Mellon University designed a set of "expert system" called xcon for digital equipment company, using artificial intelligence program.It can be simply understood as the combination of "knowledge base + inference engine"（知识库+推理机）

Second trough

In 1987, the performance of desktop computers（台式机） produced by Apple and IBM exceeded that of general-purpose computers（通用计算机） produced by manufacturers such as Symbolics.

Rise again

With the gradual development of AI technology, especially neural network technology, artificial intelligence technology began to enter a period of stable（稳定的） development.

On May 11, 1997, IBM's computer system "dark blue" defeated world chess champion Kasparov(卡斯帕罗夫).

Last three years

In the last three years, it has detonated（引爆） a business revolution. Internet giants such as Google, Microsoft and Baidu, as well as many start-up technology companies, have joined the battlefield of artificial intelligence products.

Some key technologies Turing test

Turing tests believe that if a machine can talk with humans (via telex equipment(通过电传设备)) and can not be identified by its machine identity, then the machine is intelligent. This simplification enables Turing to convincingly illustrate that the "thinking machine" is possible.

On June 8, 2014, a chat robot named Eugene Gustman（尤金·古斯特曼） successfully convinced humans that it was a 13-year-old boy and became the first computer in history to pass the Turing test. This is considered to be a milestone（里程碑） in the development of artificial intelligence.

SVM（Support Vector Machine）

这里的知识比较专业，所以我选择用汉语的方式来进行讲解，我善解人意的英语老师应该会体谅我的。

SVM，中文名意为支持向量机，而SVM作为一种分类工具，在AI届具有很高的地位，如果有想学习人工智能的同学，在只要与机器相关的考试都会考到，毕竟在Deep Learning出现（2012）之前，SVM是最优秀的分类器了

Two advantages of SVM：

1.The model has good effect（模型效果好）

2.Mathematical logic advance（数学逻辑超前）

Specific explanation

 这两种分类方式看起来作用效果相似，但是如果再加上两个点呢

 这样我们可以看见，上面那个点距离绿色区域更近，因此，由数学模型可知，上面的点更大概率归属于绿色区域，而下面这个点更大概率归红色区域。而这个结果就只有第二个分类方式做对了

因此我们可以看出有两件事保证了SVM的学霸属性：

1.尽可能保证分类的正确性

2.尽可能加大两个类别间的距离

在分类的这条界线上我们可以看出，以绿色一侧为例，当一个样本点距离界限越远，那么这个点的可信程度越高，因此想要得出最优界限，只要找到线附近的样本，使这两点到界限的距离越远越好，而这些决定线的样本叫做支持向量，这也是SVM名字的由来。

SVM是一种监督类学习的线性分类器，而我们学习的目标是找到合适的分界线，即Decision Boundary（决策分界线），而对这个超平面进行线性描述的过程所涉及了许多线性代数与核函数的知识，就不在这里推了，之后可能会写一篇blog进行推理（前提是我能完全理解），这里不在过多赘述。

有一些情况，例如，如果出现样本分布有交叉，如下图，

我们无法用一条线来进行完美的分类，因此，我们就需要给予一定的容错率，即可以允许有一些错误的分类，但是对界限的分类就更高，要求在支持向量最远时，保证错误率最小。

当数据完全不可分，即一条直线无法将数据完全区分时，我们就要进行一些处理

我们引入一个概念名为KERNEL，即，将二维的数据点，映射到更高的维度，在更高的维度，我们可以运用线性超平面去求解SVM 

或者如下图所示，

也能解决一些SVM问题。

注：以上图片来自B站up主@五分钟机器学习@KnowingAI知智

名称附上传送门，如若侵权，诚感抱歉

Boosting

太难了，我不会

附dalao链接，感兴趣的可以自己去看

boosting系列算法_jingshuiliushen_zj的博客-CSDN博客_boosting算法

Boosting方法详解_nstarLDS的博客-CSDN博客_boosting方法

我感觉dalao写的很清楚，但我无法清楚的讲解，索性不误人子弟了。

真•蒟蒻一枚

Deep Learning

Deep Learning是机器学习中一个非常接近AI的领域，其动机在于建立，模拟人脑进行分析学习的神经网络。

Depth 概念：depth: the length of the longest path from an input to an output.

Deep Architecture 的三个特点：深度不足会出现问题；人脑具有一个深度结构（每深入一层进行一次abstraction，由lower-layer的features描述而成的feature构成，就是feature hierarchy问题，而且该hierarchy是一个稀疏矩阵）；认知过程逐层进行，逐步抽象

Deep Learning Algorithm 的核心思想：

    把learning hierarchy 看做一个network，则

    ①无监督学习用于每一层网络的pre-train；

    ②每次用无监督学习只训练一层，将其训练结果作为其higher一层的输入；

    ③用监督学习去调整所有层

Deep Learning在2012年提出，在处理较多数据上，性能可以超SVM好多，而SVM又可以在性能上超出其他算法（随机森林除外），但在少数数据上，Deep Learning的表现不如SVM

不对Deep Learning进行深入的解释在这里，由于其涉及得知识点很多，包括监督学习，无监督学习，分类，回归，Sparse DBN等

如果你对Deep Learning 感兴趣，在这里推荐大牛Yoshua Bengio的综述

http://www.iro.umontreal.ca/~lisa/pointeurs/TR1312.pdf

文章

The monograph or review paper Learning Deep Architectures for AI (Foundations & Trends in Machine Learning, 2009).
The ICML 2009 Workshop on Learning Feature Hierarchies webpage has a list of references.
The LISA public wiki has a reading list and a bibliography.
Geoff Hinton has readings from last year’s NIPS tutorial.
阐述Deep learning主要思想的三篇文章： 
Hinton, G. E., Osindero, S. and Teh, Y., A fast learning algorithm for deep belief netsNeural Computation 18:1527-1554, 2006
Yoshua Bengio, Pascal Lamblin, Dan Popovici and Hugo Larochelle, Greedy Layer-Wise Training of Deep Networks, in J. Platt et al. (Eds), Advances in Neural Information Processing Systems 19 (NIPS 2006), pp. 153-160, MIT Press, 2007<比较了RBM和Auto-encoder>
Marc’Aurelio Ranzato, Christopher Poultney, Sumit Chopra and Yann LeCun Efficient Learning of Sparse Representations with an Energy-based Model, in J. Platt et al. (Eds), Advances in Neural Information Processing Systems (NIPS 2006), MIT Press, 2007<将稀疏自编码用于回旋结构(convolutional architecture)>

Language Python

Because of its simplicity and ease of use, Python is one of the most widely used programming languages in the field of artificial intelligence. It can be used seamlessly with data structures and other commonly used AI algorithms

Advantages of Python

Python是一种用Lisp和Java编译的语言，通过比较可以发现Lisp与Python非常相似，但在人工智能方面，Python有许多优势：

1.与平台无关，可以在现在每一个*nix版本上使用

2.与其他面向对象编译的语言相比，Python的学习更简单快速

3.Python有许多图像加强库像Python Imaging Libary,VTK和Maya 3D可视化工具包，Numeric Python, Scientific Python和其他很多可用工具可以于数值和科学应用。

4.Python的设计非常好，快速，坚固，可移植，可扩展。很明显这些对于人工智能应用来说都是非常重要的因素。

5.它是开源的。可以得到相同的社区支持。

.

C++

C + + is the fastest programming language in the world.

C + + is very time sensitive(敏感的), which is very useful for AI projects. For example, C + + can be widely used by search engines.

Lisp

Lisp因其出色的原型设计能力和对符号表达式的支持在AI领域崭露头角
LISP stands out in the field of AI because of its excellent prototyping ability and support for symbolic expressions.

Prolog

Prolog is comparable（可用性） to LISP in terms of availability（可用性）

Prolog一种逻辑编程语言，主要是对一些基本机制进行编程，对于AI编程十分有效
Prolog is a logic programming language, which mainly programs some basic mechanisms. It is very effective for AI programming

Java

Java也是AI项目的一个很好的选择。它是一种面向对象的编程语言，专注于提供AI项目上所需的所有高级功能，它是可移植的，并且提供了内置的垃圾回收。
Java is also a good choice for AI projects. It is an object-oriented programming language that focuses on providing all the advanced functions required on AI projects. It is portable and provides built-in garbage collection.

Example metaverse（元宇宙）

 As we all know,Zuckerberg changed the company name from Facebook to meta, and put the focus of the company's work into the development of metaverse.

AI technology is a hard technology necessary for the development of the meta universe

Others Virtual personal assistant

Such as Siri,Cortana.

Through artificial intelligence technology, they will recognize your voice, process the information and complete the instructions you give

ai人工智能开发语言_自然语言理解过程| 人工智能_cumubi7453的博客-CSDN博客

Driverless

Identify road conditions, perceive the environment, plan routes（Dijisktra）, etc

人工智能在无人驾驶中的应用_YukinoAI的博客-CSDN博客_人工智能在无人驾驶领域的应用

online Service

AI智能在线客服系统，如何实现自动回复功能?_微服网络的博客-CSDN博客

Music and movie recommendation service

Netease cloud, QQ music and so on.

Smart home device

智能家居与人工智能_我云的博客-CSDN博客_智能家居与人工智能

借鉴文章

目前人工智能技术，主要有应用于哪些领域？_多智时代的博客-CSDN博客_人工智能的应用领域

人工智能的发展历程_学习机器学习-CSDN博客_人工智能发展史

Python在人工智能中的作用_urbanvice的专栏-CSDN博客_python在人工智能领域的应用论文

机器学习——深度学习(Deep Learning)_Rachel Zhang的专栏-CSDN博客_deep learning

最适合人工智能开发的5种编程语言_Elvis技术小屋-CSDN博客_人工智能开发语言

还有许多文章无法一一列举，往作者海涵。