RNN Learning
RNN Learning[toc]
Vanilla RNN ArchitectureAbstract designRNN is the abbreviation of the Recurrent Neural Network. A piece of slice in RNN (because it is recurrent) is as below:
For each iteration, a piece of RNN model receives a current input (optional), and combine it with the previous state to send to the processing function (linear weight and non-linear activation function) to produce the new state. The new state can then used to generate the current output (optional). The hidden transition ...
How to train a good model 1
How to train a good modelA good setupTo train a good model, apart from designing an overall model structure, we should prepare suitable training components for our training to get the model converge. They includes:
Activation functions
Use ReLU for the first hand, may be good enough
data preprocessing
Depending on the tasks, different Conv Nets structure use different data preprocessing techniques.
weight initialization (Kaiming & Xavier initialization)
We can initialize the weight acc ...
CNN Architecture Dig Deeper
CNN Architecture Dig DeeperAnalogy to the fully connected layer and a single neuron
Convolution LayersDimensional description of CNN
:warning: Notice that we have a batch number of N in practice. So the input is usually of 4-dim.
Stacking of layersConv + ReLU connects the layers together
Receptive FieldsFor convolution layers, each kernels allow us to see a particular local feature with size in the input equals to the kernels size. So a 3x3 kernel can only see the feature of the local feature ...
The gradient of functions in linear classifiers in backprop
多维情况下的梯度多维情况下的梯度是比较棘手的问题，其对应的数学知识是向量值函数。
本文中分析的输入输出的维度格式为：X = (num,dim) , W = (dim, class) , Y = (num, class)其现实含义是将 num 个 dim 维样本，乘以矩阵 W，即将每个横向量样本通过一个 W 的列向量评分线性函数 $x_i * w_i + b =s_{ic_i}$ ，得到该样本在该类上的评分 $s_{ic_i}$ .
如果将一个输出 y 中的横向量看作是 $f_i$ 则其只和对应的横向量 $x_i$ 有关。对于每个 $f = (f_1,f_2, f_3 \dots , f_{class})$ ,以及 $f_{c_i} = f(x_{i1}, x_{i2}, \dots , x_{in}) = x_i * w_{c_i} + b$ , 其梯度都将会是一个矩阵 w.T, 每个元素是 $\frac{df_i}{dx_ij}$ 。但是我们还有 num 个类似的 $x_i$ 横向量, 此时就有 num 个 w.T, 这是一个四维的张量. 出了 ...
文件夹增删查改监测
文件夹增删查改监测本次任务需要监测源文件夹中发生的增删查改的情况，在出现上述操作时，将复制完或者改完的样本再到目的文件夹中（使用 shutil 库）。
使用到的第三方库有 watchdog shutil subprocess 等
主要核心代码为：
1234567891011121314151617181920212223242526272829303132333435class MyHandler(FileSystemEventHandler): def __init__(self, src_dir, dst_dir): self.src_dir = src_dir self.dst_dir = dst_dir def on_modified(self, event): if not event.is_directory: if not os.path.basename(event.src_path).startswith('.'): print(f' ...
无题
Lab3 for CS3611王梓萌 521030910015
Solution
The code named topo.py is used to create the net as above.
The screenshot of mininet is given below:
Solution
Detailed implementation is in the code. The screenshot of this chatting room is as follow:
In user and program, to avoid congestion in socket, we set the socket to no_congestion mode. To both listening the server’s message and user’s input, we use select() and monitor file descriptors for activity. If file_descriptor of our socket is modifi ...
无题
软件名称：KMS 神龙版
一开始使用的是18版的KMS，但是发现无法破解。后来下载到KMS 20版，直接可以破解。
下载地址：直接谷歌搜索 KMS 神龙版就行注意需要断开代理，这样才能打开.
同时里面有彻底关闭windows defender的方法。[见此](图文详解永久彻底的关闭win10、win11自带的windows defender杀毒功能-系统操作与应用 -亦是美网络 (yishimei.cn))
sdlsdclmskdlcaaasmmdmdd
skdjncskdjncksjdnksd
csdcsdcsd
无题
CS3611 Lab 4 王梓萌 521030910015
Problem 1
Solution
For VM1 the code should be:
123456789101112# Create bridge and add portsudo ovs-vsctl add-br br1sudo ovs-vsctl add-port br1 enp0s8# Set IP address on bridge interfacesudo ifconfig br1 192.168.56.101/24 up# Add default route for bridgesudo ip route add default via 192.168.56.1 dev br1# Build VXLAN tunnel to remote hostsudo ovs-vsctl add-port s1 vxlan0 -- set interface vxlan0 type=vxlan options:remote_ip=192.168.56.102
Similarly, for VM2 the co ...
无题
Summary for Magx
Possible questions:
What question is being studied in this article? Why is this question important?
what is its core idea?
what are the advantages of the approach proposed in this paper compared to existing work?
what do you think are the disadvantages?
What question is being studied in this article? And its importance?The overall structure of this articleOverall, this paper introduced Magx, a magnetic tracking system featuring its untethered wearability, accuracy, and power ...
BeautifulSoup 尝试
BeautifulSoup 尝试BeautifySoup中的基本对象主要有两种：Tag和beautifySoup对象。
BeautifySoup对象通常的作用是接收文件，比如将html的文本用来构建bsp对象后，才能对其做一定的操作
tagTag就是html文档中的一个标签。除了初始属性之外，tag的属性就是对应html文档中标签的属性，比如class，style属性等等。此外，tag中对属性的引用和修改使用的是python 字典的方法，这点在下图中可窥见一斑：
1234567891011121314tag['class'] = 'verybold'tag['id'] = 1tag# <blockquote class="verybold" id="1">Extremely bold</blockquote>del tag['class']del tag['id']tag# <blockquote>Extreme ...
