[AI学习笔记] 简单的AI类题目汇总.
发布时间：
目录
前置概念
其实我本意是不太想记录AI类题目来着, 我用于训练AI的电脑被单独隔离开, 与打比赛的电脑分别单独存放, 但是学着写一点东西总是好的, 万一以后需要了呢(笑).
cnn
CNN又称卷积神经网络, 算是大部分入门AI的第一课, 我记得当时是看了3b1b的视频1, 了解了个大概后自己按照一些相关博客试着用numpy搭建了一个简单的模型2, 可惜识别效率太低, 就不分享了. 3b1b的可视化做得很好, 就不过多赘述了.
如果卷积的变量是序列\(x(n)\)和\(h(n)\)，则卷积的结果为:
\[x(n) * h(n) = \sum{x(i)h(n-i)}\]卷积神经网络使用了卷积运算, 用于捕捉图像中的局部特征.
apoorvctf2025
professor_oak’s_electric_misfortune
实际上在比赛刚开始的时候我只看了这一题, 描述如下:
🔥 Welcome, Pokémon Trainer! 🔥 Hello there! Welcome to the world of Pokémon! 🌍⚡ I’m Professor Oak, the leading Pokémon researcher, and I’ve been working on an advanced Pokédex upgrade that can automatically classify Pokémon types using cutting-edge neural networks. This breakthrough could revolutionize how trainers understand their Pokémon! BUT… there’s been a slight hiccup in the lab. 😰 My Pikachu got a little too excited and—well, let’s just say a few crucial data files were shocked into oblivion. ⚡💥 Now, the entire system is scrambled, and I need your help to restore order and complete the classifier! Now.. For further details, access the PDF. Heads up!! It’s password protected. The password is the name of the pokemon Ash first captured. Good Luck Trainer !!!
题目
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
另外还有一个pdf文件, 密码为caterpie, 点开后有如下内容:
Welcome, Pokémon Trainer!
Hello there! Welcome to the world of Pokémon!
I’m Professor Oak, the leading Pokémon researcher, and I’ve been working on an advanced Pokédex upgrade that can automatically classify Pokémon types using cutting-edge neural networks. This breakthrough could revolutionize how trainers understand their Pokémon!
BUT… there’s been a slight hiccup in the lab.
My Pikachu got a little too excited and—well, let’s just say a few crucial data files were shocked into oblivion. Now, the entire system is scrambled, and I need your help to restore order and complete the classifier!
Your mission: Train a CNN model to classify Pokémon by type and help me get the Pokédex back on track. Are you up for the challenge, Trainer? Let’s GO!
Notebook link: https://www.kaggle.com/code/gl3mon/apoorvquestion Dataset link: https://www.kaggle.com/datasets/gl3mon/apoorvctf
分析
很遗憾当我得知密码是绿毛虫的时候已经找不到页面了, 所以直接简单复盘一下代码部分, 核心源码如下:
class PokemonCNN(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(4, 32, kernel_size=2)
self.bn1 = nn.BatchNorm2d(32)
self.relu = nn.ReLU()
self.pool = nn.MaxPool2d(kernel_size=3)
self.drop1 = nn.Dropout(0.25)
self.conv2 = nn.Conv2d(32, 64, kernel_size=2)
self.bn2 = nn.BatchNorm2d(64)
self.relu2 = nn.ReLU()
self.pool2 = nn.MaxPool2d(kernel_size=3)
self.drop2 = nn.Dropout(0.25)
self.conv3 = nn.Conv2d(64, 128, kernel_size=2)
self.bn3 = nn.BatchNorm2d(128)
self.relu3 = nn.ReLU()
self.pool3 = nn.MaxPool2d(kernel_size=3)
self.drop3 = nn.Dropout(0.25)
self.flatten = nn.Flatten()
self.fc1 = nn.Linear(10368, 512)
self.bn4 = nn.BatchNorm1d(512)
self.drop4 = nn.Dropout(0.5)
self.fc2 = nn.Linear(512, 18)
self.softmax = nn.Softmax()
def forward(self, x):
x = self.pool(F.relu(self.bn1(self.conv1(x))))
x = self.drop1(x)
x = self.pool(F.relu(self.bn2(self.conv2(x))))
x = self.drop2(x)
x = self.pool(F.relu(self.bn3(self.conv3(x))))
x = self.drop3(x)
x = self.flatten(x)
x = self.fc1(x)
x = self.bn4(x)
x = self.drop4(x)
x = self.fc2(x)
x = self.softmax(x)
return x
emmmm, 题目好像有点复杂了, 据描述需要使用GPU T4x2, 不过结合之前学习deepseek的一些经验, 理论上是可以加入以下代码修改的:
def __init__(self):
self.device = "cuda" if pt.cuda.is_available() else "cpu"
print(f"Using device: {self.device}")
然后先进第一题:
Question 1: What is the seed of the given code?
程序中给出了提示: I have not mentioned the seed for security reasons. But I'll let you in on a little secret. It's the pokemon number for charizard., 但是理解不了, 复盘的时候我直接上了爆破:
from EXP.pwn import *
def test(num):
io("./test")
print(r())
s(str(num).encode())
data = bool(r() == b'> Incorrect answer for question 1.\n')
close()
return data
for i in range(10000):
if test(i):
continue
else:
print(i)
break
# [*] Stopped process './test' (pid 613)
# 151
不是很理解喷火龙为什么是151总之, 但是拿到了种子编号, 可以直接修改为pt.manual_seed(151).
Question 2: What’s the number of input features of the first linear layer?
跑不出来, 直接放代码, 有时间再慢慢说.
EXP
import torch as pt
import torch.nn.functional as F
import torch.nn as nn
pt.manual_seed(151) # I have not mentioned the seed for security reasons. But I'll let you in on a little secret. It's the pokemon number for charizard.
from torch.utils.data import Dataset, DataLoader
from torchvision import datasets
from torchvision.transforms import ToTensor
import matplotlib.pyplot as plt
import os
import pandas as pd
from torchvision.io import read_image
class CTFDataset(Dataset):
def __init__(self, annotation_file, img_dir, transform = None, target_transform = None):
self.img_labels = pd.read_csv(annotation_file, index_col=0)
self.label_mapping = {
'Bug': 0, 'Dark': 1, 'Dragon': 2, 'Electric': 3, 'Fairy': 4,
'Fighting': 5, 'Fire': 6, 'Flying': 7, 'Ghost': 8, 'Grass': 9,
'Ground': 10, 'Ice': 11, 'Normal': 12, 'Poison': 13, 'Psychic': 14,
'Rock': 15, 'Steel': 16, 'Water': 17
}
self.img_labels['class'] = self.img_labels['type'].map(self.label_mapping)
self.img_dir = img_dir
self.transform = transform
self.target_transform = target_transform
def __len__(self):
return len(self.img_labels)
def __getitem__(self, idx):
img_path = os.path.join(self.img_dir, self.img_labels.iloc[idx]['type'], f'{self.img_labels.iloc[idx]["pkn"]}'+'.png')
image = None
try:
image = read_image(img_path).to(pt.float32)
except:
image = pt.zeros(4, 256, 256)
label = self.img_labels.iloc[idx]['class']
if self.transform:
image = self.transform(image)
if self.target_transform:
label = self.target_transform(label)
return image, label
ds = CTFDataset(
annotation_file = '/kaggle/input/apoorvctf/final_y.csv',
img_dir = '/kaggle/input/apoorvctf/train',
)
train_dataloader = DataLoader(ds, batch_size=4, shuffle=False)
class PokemonCNN(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(4, 32, kernel_size=2)
self.bn1 = nn.BatchNorm2d(32)
self.relu = nn.ReLU()
self.pool = nn.MaxPool2d(kernel_size=3)
self.drop1 = nn.Dropout(0.25)
self.conv2 = nn.Conv2d(32, 64, kernel_size=2)
self.bn2 = nn.BatchNorm2d(64)
self.relu2 = nn.ReLU()
self.pool2 = nn.MaxPool2d(kernel_size=3)
self.drop2 = nn.Dropout(0.25)
self.conv3 = nn.Conv2d(64, 128, kernel_size=2)
self.bn3 = nn.BatchNorm2d(128)
self.relu3 = nn.ReLU()
self.pool3 = nn.MaxPool2d(kernel_size=3)
self.drop3 = nn.Dropout(0.25)
self.flatten = nn.Flatten()
self.fc1 = nn.Linear(10368, 512)
self.bn4 = nn.BatchNorm1d(512)
self.drop4 = nn.Dropout(0.5)
self.fc2 = nn.Linear(512, 18)
self.softmax = nn.Softmax()
def forward(self, x):
x = self.pool(F.relu(self.bn1(self.conv1(x))))
x = self.drop1(x)
x = self.pool(F.relu(self.bn2(self.conv2(x))))
x = self.drop2(x)
x = self.pool(F.relu(self.bn3(self.conv3(x))))
x = self.drop3(x)
x = self.flatten(x)
x = self.fc1(x)
x = self.bn4(x)
x = self.drop4(x)
x = self.fc2(x)
x = self.softmax(x)
return x
sum(dict(pk.named_parameters())["fc1.weight"][0]) = tensor(-3.6320, grad_fn=<AddBackward0>)
dict(pk.named_parameters())["fc2.weight"][2][15] = tensor(-0.0241, grad_fn=<SelectBackward0>)
# apoorvctf{P1k4chu_0ch0t0n4_pr1nc3ps}
ghctf2025
没想到国内也有比赛单开了一个赛道, 有时间试着跟wp学习学习.
mortis
参考文献
3Blue1Brown.【官方双语】深度学习之神经网络的结构 Part 1 ver 2.0[EB/OL].bilibili.https://www.bilibili.com/video/BV1bx411M7Zx.2017-10-19 ↩
qxcheng.用numpy实现CNN卷积神经网络[EB/OL].博客园.https://www.cnblogs.com/qxcheng/p/11729773.html.2019-10-23 ↩
