蛮力识别 [英] Brute Force recognition

问题描述

您好机器学习社区，

我们尝试在我的公司(我们有一些挑战游戏)的一小段时间内使用机器学习来制作POC.

We try to make a POC using machine learning in a small time frame in my company (kind of challenge game we have). 

在这三天中，我们希望围绕一个主题发现暴力攻击"进行发现和实验机器学习.

During these three days, we would like to discover and experiment machine learning around a topic that is: "detect brute force attacks".

我们的示例是想象一个系统，您必须输入6到9位数字才能输入.

Our example is to imagine a system where you have to enter a digit code 6 to 9 digits to enter.

您可以尝试3次，如果全部失败，则必须重新连接.

You can try three times and if all failed you have to reconnect.

我们使用

trya，datetime，ishackingScenario，digitx，digity，digitz.

trya, datetime, ishackingScenario, digitx, digity, digitz.

我们创建了具有暴力破解功能的数据集(+1表示digity = digitx + 1和digitz = digity + 1). +10和我们想象的其他

We create dataset with brute force hacking (+1 meaning digity=digitx+1 and digitz=digity+1). with +10 and other we imagined

我们创建了其他一些具有有效连接或存在我们可以想象的正常错误的数据集.

We create some other dataset with valid connection or with normal errors that we can imagine.

所以我们准备好训练系统，并且能够分辨出什么时候有问题，什么时候没有.

So we are ready to train the system and we are able to tell when this is a problem and when this is not.

然后，我们希望能够发送包含有效行和无效行的csv，并希望查看系统是否能够识别出暴力模式并向我们发出警报.

Then we would like to be able to send a csv with a mix of valid and invalid lines and we would like to see if the system is able to recognize a brute force pattern and alert us.

对我们应该用来实现这一目标的方向有任何想法吗?

Any idea of the direction we should use to try to achieve this?

我们应该看看任何鼓舞人心的教程吗?

Any inspiring tutorial we should look at?

欢迎任何帮助.

谢谢大家阅读到此为止.

Thank you all for reading till there.

理查德.

推荐答案

Richard，

Hi Richard,

这似乎很有趣.据我了解，您的数据集中的观察结果已标记为标签，也就是说，出于培训目的，您知道在哪种情况下尝试是攻击或有效连接.因此，机器学习模型应该是一种受监督的模型.而且， 由于标签是两个类别，因此您应该应用分类模型. AML Studio中有很多两类分类算法，您可以训练其中的几种分类算法，并比较测试数据集的评分效果.我会尝试神经 给定您需要执行的模式识别任务，在网络中应用超参数调整.

that seems funny. As far as I understand the observations in your dataset are labelled, that is, for training purposes you know in which cases the try was an attack or a valid connection. So the machine learning model should be a supervisioned one. Moreover, as the labels are two categories, you should apply a classification model. There are lot of two-class classification algorithms in AML Studio, you can train several of them and compare the scoring performance on your test dataset. I would try with a neural network applying hyperparameter tuning, given the pattern recognition task you need to conduct.

最好

Ariel

这篇关于蛮力识别的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！