Only Truth Matters

1. 1 Introduction

1.1. 1.1 What is a Decision Tree

A tree shaped supervised learning algorithm

Problem setting:

• Set of possible instances $X$
• each instance $x$ in $X$ is a feature vector $x = < x _{ 1 } , x_ { 2 } \ldots x _ { n } >$
• Unknown target function $f : X \rightarrow Y$
• Set of function hypotheses $H = { h | h : X \rightarrow Y }$

Input:

• Training examples $\left{ < x ^ { ( i ) } , y ^ { ( i ) } > \right}$

Output:

• Hypothesis $h \in H$ that best approximates target function $f$

1. 1 Introduction

Testing whether a hypothesis is true or false by calculating the probability of an event in a prolonged experiment is known as frequentist statistics

• An experiment with an infinite number of trials guarantees $p$ with absolute accuracy
• It’s not practial to conduct an experiment with an infinite number of trials
• deciding the value of this sufficient number of trials is a challenge
• If we can determine the confidence of the estimated $p$ , it will allow us to decide whether to
• accept the conclusion
• extend the experiment with more trials until it achieves sufficient confidence
• prior beliefs (for example, coins are usually fair and the coin used is not made biased intentionally, therefore $p≈0.5$)
• play a significant role in shaping the outcome of a hypothesis test
• However, it ==can’t== be used along with frequentist statistics

2. 2 Bayesian Learning

Consider the flip coin experiment, if you flip the coin $10$ times, there are 2 case:

• 5 heads and 5 tails

• more confidence about that $p=0.5​$

• $x$ head and $10-x$ tails

Now you have 2 options

• frequentist statistics :Neglect prior beliefs, just based on data
• Bayesian Learning :Adjust your belief according to observation

2.1. 2.1 Bayes’ Theorem

Bayes’ theorem
$$
P ( \theta | X ) = \frac { P ( X | \theta ) P ( \theta ) } { P ( X ) }
$$

1. 1 Introduction

How machine can learn from data?

First of all, we need to understand human brain. And we can make the machine learning application work the same way as our brain did.

Most simple algorithms: Find-S

2. 2 What is Concept Learning

“The problem of searching through a predefined space of potential hypotheses for the hypothesis that best fits the training examples.”

​ — Tom Michell

Human learning: acquiring general concepts from past experiences.

Machine learning: find a hypothesis that best fits the training example.

Notaion:

• target concept $c$
• object $X$
• all hypothesis set $H$
• $c : X \to { 0,1 }$

1. 1 命名规则

STM32F103C8T6

• STM：基于ARM®的32位微控制器

• F：通用类型

• 103：增强型

• C：48脚

• 8：64K字节的闪存存储器

• T：LQFP封装

• 6：工业级温度范围，-40°C~85°C

2. 2 RCC

Reset and Clock Control——复位和时钟控制

2.1. 2.1 复位

三种不同方式，影响的区域不同

• 系统复位：除时钟控制器寄存器中的复位标志位和备份区域
• 上电复位：除备份区域
• 备份区域复位：只影响备份区域

2.2. 2.2 时钟

三种不同的时钟来源，用来驱动系统时钟

1. 1 Overall

共性：

1. 立即数不能作为目的操作数
2. 以A为目的操作数的指令会影响Parity
3. Rn与Rn、Rn与@Ri、@Ri与@Ri不能同时出现在指令的源、目的操作数中。

操作数的表现形式

• 内部RAM：A、Rn、@Ri、direct、#data
• 外部RAM：@DPTR、@Ri
• ROM：@A+DPTR、@A+PC

2. 2 数据传送指令

除以累加器A为目的操作数的数据传送指令对P标志位有影响外，其余数据传送指令均不影响标志位。

2.1. 2.1 内部RAM

• MOV
• 在A、Rn、@Ri、direct、#data之间互传
• 除了*Rn之间、Rn与@Ri之间、@Ri之间*
• direct可以自己传自己

2.2. 2.2 外部RAM

1. 1 Embedded System Introduction

An embedded system is a microcontroller or microprocessor based system which is designed to perform a specific task

1.1. 1.1 Architecture

• Von Neumann: data and code lie in the same memory blocks
• Harvard: data and code lie in different memory blocks

1.2. 1.2 Instruction Set

• CISC: Easy to use, higher clock cycle
• RISC: Hard to use, lower clock cycle

2. 2 8051 Introduction

In 1981, Intel introduced an 8-bit microcontroller called the 8051.

• 128 bytes of RAM
• 4K byte of on-chip ROM
• Max 64K at all because PC is 16-bit : (0000 to FFFF address)
• two timers
• 4 ports (8-bit wide)
• 3 internal and 2 external Interrupts
• Family Member
• 8052: 8K ROM, 256 byte RAM, 3 Timer
• 8031: 0K ROM, 128 byte RAM, 2 Timer

2.1. 2.1 Registers

1. The Big Picture

taught in MIT

1.1. Row Space = $C(A^T)$

linear combination, fill the plane
$$
A=\left[\begin{array}{lll}
1 & 2 & 3 \
4 & 5 & 6
\end{array}\right]
$$
Only 2 rows, so can’t fill the 3D, here we introduce null space

1.2. Null Space = $N(A)$

$$
A v=\left[\begin{array}{l}
0 \
0
\end{array}\right]
$$

In this situation
$$
v=\left[\begin{array}{l}
1 \
-2 \
1
\end{array}\right]
$$
is the null vector, prependicular line

1.3. Column Space = $C(A)$

The column space the whole space

1.4. Left Space =$N(A^T)$

1. 1 Introduction

A process that transforms one or more input images into an output image.

Main purpose: enhance an image for human viewing

A image is just a matrix.

• Value: uint8 ,[0-255] (from darkest to brightest)
• if use complex algorithms, may use floating-point numbers
• Descibe: width $\times$ height

2. 2 Useful Algorithms

Histogram: the number of times each pixel value occurs.

If a picture was under-exposed, the histogram would shift to the left.

Find peak in nearby x value:

3. 3 Monadic & Diadic Operations

1. 1 Introduction

The light that reaches the eye,or the camera, is a function of the illumination impinging on the scene and the material property known as reflectivity.

2. 2 Light

Each color is a single frequency or wavelength of electro-magnetic radiation.We perceive the wavelengths between 400 and 700 nm as different colors.

In general, the light we observe can be represented as a function $E(\lambda)$ ,$\lambda$ is the wavelength. This is a function about **power **related to

• source temperature $T$ (e.g. 4700K)
• Planck’s constant $h$ , Boltzmann’s constant $k$, speed of light $c$
• ……

2.1. 2.1 Absorption

Spectrum will be changed (be absorbed)

1. 1 Introduction

SLAM builds a map and localize the sensor in the map with a strong focus on real-time operation.

Camera: cheap and provide rich information of the environment.

• Monocular camera, cheapest and smallest camera
• depth is not observable
• scale drift and mail fail if performing pure rotations
• RGB-D camera, all these issue can be solved.
• Outdoor performance is not good. Usually used in indoor environment

1.1. 1.1 Main Idea

At its heart, SLAM is an optimization problem, where the goal is to compute the best configuration of camera poses and point positions in order to minimize reprojection error (the difference between a point’s tracked location and where it is expected to be given the camera pose estimate, over all points). —from Kudan

The optimization method: Bundle adjustment, iteratively approaches the minimum error for the whole system.

• Problem: time consuming to find the best solution
• But with the help of multi-core machine, this problem was solved

Another essential technique: relocalization

1.2. 1.2 How it Works