-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
1 parent
19ea6b8
commit b70c7b4
Showing
30 changed files
with
926 additions
and
126 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,85 @@ | ||
# Dive into deep learning | ||
|
||
## 0. Installation | ||
|
||
- Python 3.12 安装 NumPy 有 Bug,降级到 3.11。 | ||
|
||
## 1. Intro | ||
|
||
- 机器学习是从经验中学习的算法。 | ||
|
||
需要明白以下概念的含义: | ||
|
||
- 监督、无监督、自监督、强化学习 | ||
- 监督学习所解决的:回归、分类、标注、搜索、推荐、序列学习问题 | ||
|
||
本章还介绍了最新的人工智能发展趋势,如深度学习、自然语言处理、计算机视觉、强化学习、生成对抗网络等。 | ||
|
||
!!! note "交叉:计算机体系结构" | ||
|
||
本章也对人工智能和计算机计算能力的发展进行了比较。值得注意的是,RAM 的增长速度落后于数据集和计算能力的增长速度,因此现在的统计模型应当更加内存高效,也就是在访存期间做更多计算。作者提到,这也是多层感知机、CNN 等模型在上个世纪就已经被提出,但进来才得以广泛应用的原因之一。 | ||
|
||
## 2. Preliminaries | ||
|
||
### Data Manipulation | ||
|
||
!!! note "`tensor`" | ||
|
||
创建 | ||
|
||
```python | ||
torch.arange(12, dtype=torch.float32) | ||
torch.zeros((2, 3, 4)) | ||
torch.ones((2, 3, 4)) | ||
torch.randn(3, 4) | ||
torch.tensor([[2, 1, 4, 3], [1, 2, 3, 4], [4, 3, 2, 1]]) | ||
``` | ||
|
||
属性: | ||
|
||
```python | ||
x.shape | ||
``` | ||
|
||
方法: | ||
|
||
```python | ||
x.reshape(3, 4) | ||
x.reshape(-1, 4) # auto infer | ||
``` | ||
|
||
操作: | ||
|
||
```python | ||
# 一元、二元运算都是 element-wise 的。 | ||
torch.exp(x) | ||
x + y, x - y, x * y, x / y, x ** y | ||
x == y, x < y, x > y | ||
# 其他 | ||
torch.cat((x, y), dim=0) | ||
x.sum() | ||
``` | ||
|
||
广播: | ||
|
||
- 对于长度为 1 的维度,拷贝使得两个张量的维度相同。 | ||
- 执行 element-wise 运算。 | ||
|
||
|
||
!!! note "其他" | ||
|
||
```python | ||
id() | ||
``` | ||
|
||
## 3. Linear Neural Networks | ||
|
||
### Data Preprocessing | ||
|
||
### Linear Algebra | ||
|
||
### Calculus | ||
|
||
### Automatic Differentiation | ||
|
||
### Probability and Statistics |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,39 @@ | ||
# 人工智能综述 | ||
|
||
[AI Expert Roadmap](https://github.com/AMAI-GmbH/AI-Expert-Roadmap) | ||
|
||
在具体地进入人工智能的学习前,我们先来看看人工智能的发展历程,这对了解人工智能的发展脉络和其中的一些重要概念有很大帮助。 | ||
|
||
!!! note "知识库方法 Knowledge Base" | ||
|
||
将关于世界的知识用形式化的语言进行硬编码。计算机使用逻辑推理规则来自动理解这些形式化语言中的声明。 | ||
|
||
人工智能早期解决对人类智力来说非常困难,但对计算机来说相对简单的问题,比如抽象和形式化的任务。例子有:IBM 的深蓝国际象棋系统。 | ||
|
||
知识库方法遇到了困难,因为难以设计出足够复杂的形式化规则来精确地描述世界。最著名的知识库项目 Cyc 不能理解人在早上剃胡须:它知道人体的构成不含有电器零件,但由于人拿着剃须刀,它认为实体含有电器部件。因此它产生疑问:人在刮胡子时是否仍是一个人? | ||
|
||
!!! note "机器学习 Machine Learning" | ||
|
||
系统具备自己获取知识的能力,能够从原始数据中提取模式。 | ||
|
||
主要使用数理统计方法。 | ||
|
||
由人从数据中提取特征,将数据处理成结构化的形式,然后让机器学习算法从中提取模式。 | ||
|
||
但是,有很多任务我们很难知道如何提取特征。比如检测照片中的车,使用轮子作为特征吗?但图像可能因场景而异,轮子可能被遮挡,或者车可能是在水中。 | ||
|
||
!!! note "表示学习 Representation Learning" | ||
|
||
用机器学习来发掘表示本身,学习到的表示往往比手动设计的好。 | ||
|
||
!!! note "自编码器 Autoencoder" | ||
|
||
由编码器函数和解码器函数组成,编码器将输入数据映射到表示空间,解码器将表示空间映射回原始数据空间。期望是输入数据经过编解码器后尽可能多地保留信息,同时希望新的表示有各种好的特性。 | ||
|
||
!!! note "深度学习 Deep Learning" | ||
|
||
一种机器学习技术,它使用神经网络来学习数据的表示形式。 | ||
|
||
!!! note "前馈深度网络/多层感知机 Multilayer Perceptron" | ||
|
||
一种最简单的神经网络,由多个神经元组成的多层结构。 |
File renamed without changes.
File renamed without changes.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,25 @@ | ||
# 草稿 | ||
|
||
## 文献综述 | ||
|
||
!!! info "[东南大学图书馆:如何写好文献综述](http://www.lib.seu.edu.cn/upload_files/file/20220523/_20220523153114.pdf)" | ||
|
||
### 内容 | ||
|
||
文献综述的内容一般包括该研究领域的: | ||
|
||
- **研究现状**:包括主要学术观点、前人研究成果和研究水平、争论焦点、存在的问题及可能的原因等。 | ||
- **发展趋势**:新水平、新动态、新技术、新发现、发展前景等。 | ||
|
||
文献综述需要对以上内容进行**行综合分析、归纳整理和评论**,并提出**自己的见解和研究思路**。 | ||
|
||
文献综述的目的是帮助读者确认该论文所研究的问题**与以往同类或同领域论文相比较所具有的价值**及在选题或研究内容与方法上**是否具有创新性或新的进展**。 | ||
|
||
### 结构 | ||
|
||
按时间顺序、研究主题、研究方法、学术流派等。 | ||
|
||
- 时间顺序:最早研究该主题的文献是什么?这个研究领域随着时间的推移是如何变化的(以及为什么)?最新的发现是什么? | ||
- 研究主题:研究人员使用的中心主题和类别是什么?有哪些证据来证明这些主题? | ||
- 研究方法:哪些方法在这个领域已经被利用?哪种方法是最受欢迎的(以及为什么)?各种方法的优缺点是什么?现有方法如何为我的研究提供参考? | ||
- 学术流派:已有研究的主要学术流派和观点是什么?观点之间存在怎样的逻辑关系? |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,31 @@ | ||
# 多量子比特和量子纠缠 | ||
|
||
## 张量积 | ||
|
||
## 纠缠态和 EPR 佯谬 | ||
|
||
## Bell 态中的信息 | ||
|
||
- PPT:信息分布在两个量子比特之间(not local),任意单个量子比特不提供任何信息。 | ||
|
||
要理解上面这一描述,课上采用计算期望值 $\braket{L}$ 的方法。 | ||
|
||
1. $L$ 的特征值 $\lambda$,特征向量 $\ket{\lambda}$ | ||
2. 把量子态分解到特征向量上 $\ket{\psi} = \sum_\lambda \ket{\lambda}\braket{\lambda|\psi}$,再应用 $L$ | ||
3. 再测量,得到期待值 $\braket{L} = \sum_\lambda \lambda \mathrm{P}(\lambda)$ | ||
|
||
!!! note "牢记:$\mathrm{P}(\lambda) = \braket{\psi|\lambda}\braket{\lambda|\psi}$" | ||
|
||
本质上就是带权(概率)的特征值求和。 | ||
|
||
测量 Bell 态 $\ket{\psi} = \frac1{\sqrt{2}}(\ket{01}+\ket{10})$ 的第一个比特,我们对第一个比特应用由泡利矩阵生成的任意酉矩阵: | ||
|
||
$$ | ||
(\vec{\sigma}^{(1)} \cdot \vec{n}) \otimes I | ||
$$ | ||
|
||
计算它的期望值: | ||
|
||
|
||
|
||
## 密度矩阵 |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,19 @@ | ||
# 量子系统随时间的演化 | ||
|
||
本节课在线性代数上加强了难度。让我们再次复习相关概念。 | ||
|
||
???+ note "Hermitian Operators" | ||
|
||
定义: | ||
|
||
- 自伴(共轭转置等于本身):$A^\dagger = A$。 | ||
- 伴随矩阵 $A^\dagger$ 定义为:$\braket{A^\dagger|g}= \braket{f|Ag}$。 | ||
|
||
性质: | ||
|
||
- 本征值都是实数。 | ||
- 位置、动量和能量等物理量是实数,因为它们都是 Hermitian 算子的本征值。 | ||
- 因此期待值也是实数。 | ||
- 结合内积的共轭对称性质,可以得到:$\braket{A} = \braket{\psi|A\psi} = \braket{A\psi|\psi} = \braket{\psi|A\psi}^* = \braket{A}^*$。 | ||
- 本征向量正交。 | ||
|
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,5 @@ | ||
# 量子电路 | ||
|
||
!!! note "不可克隆定理" | ||
|
||
无法完美复制一个未知的量子态。 |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,4 @@ | ||
# temp | ||
|
||
## 倒排索引 Inverted Index | ||
|
Oops, something went wrong.