---
description: 发布于 2023-12-26
categories:
- study
date: 2023-12-26
slug: introduction-to-artificial-intelligence
title: 人工智能导论 极速复习版
updated: 
tags: 
- study
- artificial-intelligence
copyright: false 
---

# 人工智能导论 极速版

zstu 浙江理工大学 2023学年第1学期 人工智能 开卷考试 期末复习

## 介绍

智力是学习、理解或处理新情况或困难情况的能力；
应用知识来操纵环境或抽象思考的能力

![image-20231225234617578](https://media.opennet.top/i/2023/12/25/12paq94-0.png)

![image-20231225234637339](https://media.opennet.top/i/2023/12/25/12pem5g-0.png)

## 知识表示

### 命题逻辑

#### 理论

命题是一个**陈述句**，并且必须能判断真假

![image-20231225235102921](https://media.opennet.top/i/2023/12/25/12s6bb9-0.png)

![image-20231225235151513](https://media.opennet.top/i/2023/12/25/12sgrj8-0.png)

![image-20231225235331160](https://media.opennet.top/i/2023/12/25/12tjh9h-0.png)

![image-20231225235354291](https://media.opennet.top/i/2023/12/25/12to8n9-0.png)

![扫描件_等值式_1](https://media.opennet.top/i/2023/06/16/648bfe9e965de.jpg)

#### 题目

![image-20231225235503519](https://media.opennet.top/i/2023/12/25/12uk6uk-0.png)

![image-20231225235514757](https://media.opennet.top/i/2023/12/25/12ummri-0.png)

![image-20231225235539004](https://media.opennet.top/i/2023/12/25/12urokh-0.png)

![image-20231225235635468](https://media.opennet.top/i/2023/12/25/12vcgi0-0.png)

![image-20231225235640865](https://media.opennet.top/i/2023/12/25/12vdkgd-0.png)

![image-20231225235656649](https://media.opennet.top/i/2023/12/25/12vgzj1-0.png)

![image-20231226000723498](https://media.opennet.top/i/2023/12/26/qylt-0.png)

### 谓词逻辑

#### 理论

谓词描述关系

P(x) 或 M(x) 表示一个一元谓词逻辑

![image-20231225235825351](https://media.opennet.top/i/2023/12/25/12wh7b7-0.png)

![扫描件_定义2_3量词是描述个体常项与个体变项之_1](https://media.opennet.top/i/2023/06/16/648c0916a5990.jpg)

![扫描件_第2章一阶逻辑_1](https://media.opennet.top/i/2023/06/16/648c0a790c6d9.jpg)

量词不能随意调换顺序

量词的优先级比逻辑联结词高

![扫描件_定理2-2否定等值式_1](https://media.opennet.top/i/2023/06/16/648c1e702d20e.jpg)

![扫描件_定理24量词分配律_1](https://media.opennet.top/i/2023/06/16/648c1e87b2223.jpg)

![扫描件_一个谓词公式可以演算成与之等值的标准形式_1](https://media.opennet.top/i/2023/06/16/648c1f5964762.jpg)

前束范式运算前先换名

![image-20231226000606035](https://media.opennet.top/i/2023/12/26/1kxv-0.png)

#### 题目

![image-20231226000112663](https://media.opennet.top/i/2023/12/25/12y6z9e-0.png)

![image-20231226000120865](https://media.opennet.top/i/2023/12/25/12y8mhv-0.png)

![image-20231226000131343](https://media.opennet.top/i/2023/12/25/12yb2f5-0.png)

![image-20231226000501607](https://media.opennet.top/i/2023/12/25/130i29r-0.png)

![image-20231226000527687](https://media.opennet.top/i/2023/12/25/130nl9i-0.png)

![image-20231226000547183](https://media.opennet.top/i/2023/12/25/130ryz4-0.png)

![image-20231226000741956](https://media.opennet.top/i/2023/12/26/uu0j-0.png)

### 产生式系统

![image-20231226000812764](https://media.opennet.top/i/2023/12/26/19wws-0.png)

![image-20231226000818722](https://media.opennet.top/i/2023/12/26/1ci0x-0.png)

![image-20231226000826856](../../../../../Library/Application%20Support/typora-user-images/image-20231226000826856.png)

![image-20231226000838242](https://media.opennet.top/i/2023/12/26/1fr4e-0.png)

![image-20231226000859149](https://media.opennet.top/i/2023/12/26/1so6x-0.png)

### 框架系统

![image-20231226000924456](https://media.opennet.top/i/2023/12/26/1xyql-0.png)

![image-20231226000933405](https://media.opennet.top/i/2023/12/26/1zu7g-0.png)

![image-20231226000939683](https://media.opennet.top/i/2023/12/26/219cj-0.png)

![image-20231226000946606](https://media.opennet.top/i/2023/12/26/22s2e-0.png)

![image-20231226000957744](https://media.opennet.top/i/2023/12/26/2515e-0.png)

### 状态空间系统

![image-20231226001023321](https://media.opennet.top/i/2023/12/26/2j530-0.png)

![image-20231226001036378](https://media.opennet.top/i/2023/12/26/2lzvh-0.png)

![image-20231226001044195](https://media.opennet.top/i/2023/12/26/2nr0h-0.png)

![](https://media.opennet.top/i/2023/12/26/32bde-0.png)

### 知识图谱*

![image-20231226001135978](https://media.opennet.top/i/2023/12/26/37hzt-0.png)

![image-20231226001147128](https://media.opennet.top/i/2023/12/26/39rfc-0.png)

![image-20231226001153868](https://media.opennet.top/i/2023/12/26/3baiv-0.png)

![image-20231226001200587](https://media.opennet.top/i/2023/12/26/3l8gc-0.png)

![image-20231226001207587](https://media.opennet.top/i/2023/12/26/3mor9-0.png)

![image-20231226001215010](https://media.opennet.top/i/2023/12/26/3oayl-0.png)

## 搜索

### 理论

![image-20231226001241015](https://media.opennet.top/i/2023/12/26/3tq9x-0.png)

![image-20231226001335013](https://media.opennet.top/i/2023/12/26/4dxxz-0.png)

![image-20231226001349174](https://media.opennet.top/i/2023/12/26/4gsmw-0.png)

![image-20231226001403400](https://media.opennet.top/i/2023/12/26/4slah-0.png)

![image-20231226001410395](https://media.opennet.top/i/2023/12/26/4ubm2-0.png)

![image-20231226001434338](https://media.opennet.top/i/2023/12/26/4zdse-0.png)

![image-20231226001443568](https://media.opennet.top/i/2023/12/26/5168u-0.png)

![image-20231226001531516](https://media.opennet.top/i/2023/12/26/5k3k9-0.png)

### 题目

![image-20231226001457876](https://media.opennet.top/i/2023/12/26/547f2-0.png)

![image-20231226001507819](https://media.opennet.top/i/2023/12/26/5ev5x-0.png)

## 模型评估和选择

### 准确率 误差 过拟合

![image-20231226001623609](https://media.opennet.top/i/2023/12/26/63uiv-0.png)

![image-20231226001657563](https://media.opennet.top/i/2023/12/26/6b1ne-0.png)

![image-20231226001705605](https://media.opennet.top/i/2023/12/26/6laq6-0.png)

![image-20231226001727422](https://media.opennet.top/i/2023/12/26/6q1z1-0.png)

![image-20231226001738210](https://media.opennet.top/i/2023/12/26/6shlk-0.png)

### 评估方法

![image-20231226001941537](https://media.opennet.top/i/2023/12/26/7zxbb-0.png)

![image-20231226001952193](https://media.opennet.top/i/2023/12/26/82al5-0.png)

![image-20231226002007852](https://media.opennet.top/i/2023/12/26/8e38d-0.png)

![image-20231226002022812](https://media.opennet.top/i/2023/12/26/8haeu-0.png)

### 性能指标

![image-20231226002123085](https://media.opennet.top/i/2023/12/26/92s5r-0.png)

![image-20231226002130184](https://media.opennet.top/i/2023/12/26/94b8p-0.png)

![image-20231226002143936](https://media.opennet.top/i/2023/12/26/9793b-0.png)

![image-20231226002152418](https://media.opennet.top/i/2023/12/26/999h1-0.png)

![image-20231226002207073](https://media.opennet.top/i/2023/12/26/9kwxm-0.png)

![image-20231226002217110](https://media.opennet.top/i/2023/12/26/9mqdy-0.png)

![image-20231226002224463](https://media.opennet.top/i/2023/12/26/9ogn1-0.png)

### 题目

![image-20231226002258774](../../../../../Library/Application%20Support/typora-user-images/image-20231226002258774.png)

![image-20231226002311319](https://media.opennet.top/i/2023/12/26/a7bjt-0.png)

![image-20231226002330301](https://media.opennet.top/i/2023/12/26/aatze-0.png)

![image-20231226002403831](https://media.opennet.top/i/2023/12/26/aqphs-0.png)

## 机器学习

### 监督学习

#### 回归

![image-20231226002538556](https://media.opennet.top/i/2023/12/26/bjmlq-0.png)

#### 线性回归

![image-20231226002546750](https://media.opennet.top/i/2023/12/26/bl5o3-0.png)

![](https://media.opennet.top/i/2023/12/26/bxd9d-0.png)

![image-20231226002623421](https://media.opennet.top/i/2023/12/26/c1vgy-0.png)

![image-20231226002639972](https://media.opennet.top/i/2023/12/26/c5ehc-0.png)

#### 逻辑回归

![image-20231226002702076](https://media.opennet.top/i/2023/12/26/ciqgl-0.png)

![image-20231226002755611](https://media.opennet.top/i/2023/12/26/cu0g2-0.png)

#### 分类

![image-20231226002821957](https://media.opennet.top/i/2023/12/26/d8ejc-0.png)

#### 最近邻

![image-20231226002915035](https://media.opennet.top/i/2023/12/26/dscl1-0.png)

![image-20231226002929121](https://media.opennet.top/i/2023/12/26/dvc0r-0.png)

#### ϵ-ball 最近邻

![image-20231226003016883](https://media.opennet.top/i/2023/12/26/ee2qs-0.png)

#### K 近邻

![image-20231226003032811](https://media.opennet.top/i/2023/12/26/ehn2m-0.png)

确定 k 近邻算法（k-NN）中 k 值的大小是一个重要决策，因为它直接影响到算法的性能。没有固定的规则来选择最佳的 k 值，但是可以通过以下方法来确定：

1. **交叉验证**：最常用的方法是通过交叉验证，特别是 k 折交叉验证。在这种方法中，数据集被分成 k 个小组（folds）。算法在 k-1 个小组上训练，并在剩下的一组上测试。这个过程重复进行，每次选择不同的组作为测试集，然后取平均误差率。通过比较不同 k 值的误差率，可以选择最佳的 k 值。

2. **误差率**：对于分类问题，可以计算不同 k 值对应的误差率。误差率最低的 k 值通常会被选择。对于回归问题，可以计算均方误差（MSE）。

3. **启发式方法**：通常，k 值的选择应该是一个奇数（如果类别数为偶数），以避免决策的平局。一个常见的启发式方法是选择 \(\sqrt{n}\)，其中 n 是训练样本的数量。

4. **距离权重**：考虑距离权重可以减少更远邻居的影响，这样可以在考虑更多的邻居（较大的 k 值）的同时减少噪声数据的影响。

5. **问题特定知识**：有时，对问题的了解可以帮助确定 k 的值。例如，在高度不平衡的数据集中，较大的 k 值可能有助于防止算法过分偏向多数类。

6. **模型复杂度**：较小的 k 值会导致模型复杂度高，可能过拟合数据；较大的 k 值会导致模型简单，可能无法捕捉数据结构。因此，需要找到一个平衡点。

7. **可视化工具**：有时候，将不同的 k 值的效果可视化，例如通过绘制误差率和 k 值的关系图，可以帮助选择一个好的 k 值。

8. **规则化方法**：当数据集非常大时，可以使用规则化方法来选择 k。例如，可以将 k 设置为训练样本数量的一个小百分比。

#### 距离

![image-20231226003831748](https://media.opennet.top/i/2023/12/26/j8pc4-0.png)

#### 最近子空间分类器*

![image-20231226003846351](https://media.opennet.top/i/2023/12/26/jbsj3-0.png)

![image-20231226003910874](https://media.opennet.top/i/2023/12/26/jpt0v-0.png)

![image-20231226003919381](https://media.opennet.top/i/2023/12/26/jrj2h-0.png)

### 非监督学习

#### K-均值聚类

![image-20231226093237982](https://media.opennet.top/i/2023/12/26/fbyf07-0.png)

![image-20231226093326117](https://media.opennet.top/i/2023/12/26/fc1vsd-0.png)

![image-20231226093356816](https://media.opennet.top/i/2023/12/26/fc8ddj-0.png)

![image-20231226093406866](https://media.opennet.top/i/2023/12/26/fcj9ah-0.png)

![image-20231226093416072](https://media.opennet.top/i/2023/12/26/fcl3ih-0.png)

![image-20231226093425875](https://media.opennet.top/i/2023/12/26/fcn1sl-0.png)

![image-20231226093438871](https://media.opennet.top/i/2023/12/26/fcpv50-0.png)

![image-20231226093447604](https://media.opennet.top/i/2023/12/26/fcrvxs-0.png)

![image-20231226093504110](https://media.opennet.top/i/2023/12/26/fd3zev-0.png)

![image-20231226093608889](https://media.opennet.top/i/2023/12/26/fdq88w-0.png)

![image-20231226093621440](https://media.opennet.top/i/2023/12/26/fdt25a-0.png)

#### 谱聚类*

![image-20231226093725010](https://media.opennet.top/i/2023/12/26/fefeq5-0.png)

![image-20231226093734827](https://media.opennet.top/i/2023/12/26/fehive-0.png)

![image-20231226093748777](https://media.opennet.top/i/2023/12/26/fek8hj-0.png)

#### 题目

![image-20231226093521977](https://media.opennet.top/i/2023/12/26/fd7xv1-0.png)

![image-20231226093527701](https://media.opennet.top/i/2023/12/26/fd94np-0.png)

![image-20231226093534540](https://media.opennet.top/i/2023/12/26/fdafc2-0.png)

![image-20231226093540608](https://media.opennet.top/i/2023/12/26/fdbnkf-0.png)

![image-20231226093547420](https://media.opennet.top/i/2023/12/26/fddbk3-0.png)

#### 表示学习-线性编码*

![image-20231226094006456](https://media.opennet.top/i/2023/12/26/fg3spp-0.png)

![image-20231226094047128](https://media.opennet.top/i/2023/12/26/fgcd4u-0.png)

#### 表示学习-PCA*

![image-20231226094141936](https://media.opennet.top/i/2023/12/26/fgwpwp-0.png)

![image-20231226094150296](https://media.opennet.top/i/2023/12/26/fgymho-0.png)

表示学习应用：人脸图像压缩 图像去模糊 图像去噪 形态成分分析 图像修复

### 强化学习

![image-20231226094741449](https://media.opennet.top/i/2023/12/26/fkh64r-0.png)

![image-20231226094827747](https://media.opennet.top/i/2023/12/26/fkzvo6-0.png)

![image-20231226094841544](https://media.opennet.top/i/2023/12/26/fl2lmd-0.png)

![image-20231226094856752](https://media.opennet.top/i/2023/12/26/fl5y93-0.png)

#### Q-Learning

![image-20231226094941750](https://media.opennet.top/i/2023/12/26/flnxiq-0.png)

![image-20231226095002202](https://media.opennet.top/i/2023/12/26/flsisw-0.png)

![image-20231226095012577](https://media.opennet.top/i/2023/12/26/fm377r-0.png)

![image-20231226095028712](https://media.opennet.top/i/2023/12/26/fm6uqu-0.png)

![image-20231226100016133](../../../../../Library/Application%20Support/typora-user-images/image-20231226100016133.png)

![image-20231226100027876](https://media.opennet.top/i/2023/12/26/fs4z6n-0.png)

![image-20231226100059748](https://media.opennet.top/i/2023/12/26/fsbm9a-0.png)

## 神经网络

### 历史

![image-20231226100133188](https://media.opennet.top/i/2023/12/26/fsrfm8-0.png)

### 神经元

![image-20231226100201642](https://media.opennet.top/i/2023/12/26/fsxkl5-0.png)

### 激活函数

![image-20231226100219893](https://media.opennet.top/i/2023/12/26/fta4or-0.png)

![image-20231226100230596](https://media.opennet.top/i/2023/12/26/ftc8rj-0.png)

![image-20231226100238240](https://media.opennet.top/i/2023/12/26/ftdy3c-0.png)

![image-20231226100302971](https://media.opennet.top/i/2023/12/26/ftj4f9-0.png)

### 题目

![image-20231226100345367](https://media.opennet.top/i/2023/12/26/fu16ax-0.png)

### 前馈神经网络 FNN

![image-20231226100518069](https://media.opennet.top/i/2023/12/26/fv219k-0.png)

![image-20231226100609092](https://media.opennet.top/i/2023/12/26/gjelge-0.png)

### 循环神经网络 RNN

![image-20231226100701697](https://media.opennet.top/i/2023/12/26/gjpybv-0.png)

![image-20231226100711886](https://media.opennet.top/i/2023/12/26/gk0q3q-0.png)

![image-20231226100721630](https://media.opennet.top/i/2023/12/26/gk2ofo-0.png)

![image-20231226100735285](https://media.opennet.top/i/2023/12/26/gk5wva-0.png)

![image-20231226100840101](https://media.opennet.top/i/2023/12/26/gks7hx-0.png)

### 题目

![image-20231226101427932](https://media.opennet.top/i/2023/12/26/goa8lq-0.png)

### 损失函数

![image-20231226100924698](https://media.opennet.top/i/2023/12/26/glaggt-0.png)

![image-20231226100933684](https://media.opennet.top/i/2023/12/26/glceg0-0.png)

![image-20231226100941780](https://media.opennet.top/i/2023/12/26/gle0jw-0.png)

### 最速下降法

![image-20231226101047508](https://media.opennet.top/i/2023/12/26/gm0t5q-0.png)

![image-20231226101056401](https://media.opennet.top/i/2023/12/26/gm2o0o-0.png)

### 深度学习

![image-20231226101112828](https://media.opennet.top/i/2023/12/26/gmenkg-0.png)

![image-20231226101126707](https://media.opennet.top/i/2023/12/26/gmhov0-0.png)

### 反向传播

![image-20231226101145588](https://media.opennet.top/i/2023/12/26/gmltb5-0.png)

![image-20231226101153653](https://media.opennet.top/i/2023/12/26/gmnkf3-0.png)

![image-20231226101201752](https://media.opennet.top/i/2023/12/26/gmp5mc-0.png)

![image-20231226101209217](https://media.opennet.top/i/2023/12/26/gmz6xd-0.png)

![image-20231226101216893](https://media.opennet.top/i/2023/12/26/gn0xqw-0.png)

![image-20231226101223638](https://media.opennet.top/i/2023/12/26/gn2e31-0.png)

![image-20231226101233192](https://media.opennet.top/i/2023/12/26/gn4o1m-0.png)

### 卷积神经网络

#### 历史

![image-20231226101313011](https://media.opennet.top/i/2023/12/26/gnllxm-0.png)

#### 感受野

![image-20231226101518314](https://media.opennet.top/i/2023/12/26/gotox5-0.png)

#### 特征图

![image-20231226101648433](https://media.opennet.top/i/2023/12/26/gplk16-0.png)

#### 卷积

![image-20231226101703524](https://media.opennet.top/i/2023/12/26/gpxbt2-0.png)

![image-20231226101758482](https://media.opennet.top/i/2023/12/26/gq97fp-0.png)

#### 题目

![image-20231226101728230](https://media.opennet.top/i/2023/12/26/gq2qta-0.png)

![image-20231226101737357](https://media.opennet.top/i/2023/12/26/gq4iv6-0.png)

![image-20231226101824971](https://media.opennet.top/i/2023/12/26/gqn77h-0.png)

#### 池化

![image-20231226101853645](https://media.opennet.top/i/2023/12/26/gqtjj4-0.png)

#### 题目

![image-20231226101908468](https://media.opennet.top/i/2023/12/26/gr52q6-0.png)

#### 成就和应用

![image-20231226102025438](https://media.opennet.top/i/2023/12/26/grub4r-0.png)

![image-20231226102032190](https://media.opennet.top/i/2023/12/26/grw0po-0.png)

![image-20231226102038536](https://media.opennet.top/i/2023/12/26/grx65g-0.png)

![image-20231226102044606](https://media.opennet.top/i/2023/12/26/gryxmd-0.png)

![image-20231226102052312](https://media.opennet.top/i/2023/12/26/gs03pb-0.png)

#### 题目

![image-20231226102113742](https://media.opennet.top/i/2023/12/26/gsdblr-0.png)

![image-20231226102120925](https://media.opennet.top/i/2023/12/26/gsep5o-0.png)

### 序列学习*

序列学习（Sequence Learning）是机器学习中的一个重要分支，专注于从序列数据中学习模式和结构。序列数据是指数据元素按特定顺序排列的数据，例如时间序列数据、文本、语音或者视频帧。序列学习的目标是理解这些数据的内在结构，并能够预测或生成新的序列。

主要类型

1. **时间序列预测**：用于预测未来的数据点，例如股票价格预测、天气预报等。

2. **序列生成**：生成与训练数据类似的新序列，如文本生成、音乐创作等。

3. **序列分类**：将序列数据分类到不同类别中，例如语音识别或情感分析。

4. **序列标注**：在序列的每个元素上进行分类，常用于诸如命名实体识别、词性标注等任务。

关键技术

1. **递归神经网络（RNN）**：一种特别适用于序列数据的神经网络结构，能够处理任意长度的序列。

2. **长短期记忆网络（LSTM）**：RNN的一种变体，通过特殊的结构解决了RNN长期依赖问题。

3. **门控循环单元（GRU）**：与LSTM类似，但结构更简单，效果也很好。

4. **Transformer**：一种基于自注意力机制的模型，特别适用于大规模序列数据处理，如机器翻译。

5. **卷积神经网络（CNN）**：虽主要用于图像处理，但也可用于处理序列数据，特别是在处理较短的序列时。

应用场景

- **自然语言处理**：如机器翻译、文本摘要、情感分析等。
- **语音识别**：将语音信号转换成文字。
- **视频分析**：从视频序列中识别特定事件或对象。
- **股票市场分析**：预测股票价格变化。
- **生物信息学**：如蛋白质结构预测。

挑战

- **长期依赖问题**：在处理长序列时，传统模型（如RNN）难以记住早期信息。
- **资源消耗**：特别是对于大规模模型（如Transformer），需要大量的计算资源。
- **数据质量和可用性**：高质量、标注良好的序列数据并不总是可用。

序列学习是一个快速发展的领域，随着新技术的出现，其应用范围和效果都在不断提升。