Azure机器学习功能问题 [英] Azure machine learning capabilities questions
问题描述
我在Azure机器学习服务上询问了这个问题，建议将我的问题重定向到AML工作室论坛。
Hi ，
我正在为我的硕士论文进行机器学习（ML）实验。 我被告知我的学校可能是
能够支持我租用一些云计算时间Azure看起来非常有前景，但我有疑问。
首先概述我的项目。 ML是一个热门话题，在很多领域我都在关注网络安全，特别是
数据泄露。 我有一个数据集，来自模拟网络的大约50千兆字节的数据包捕获（PCAP）数据，在云解决方案上处理这个数据的
比构建桌面更快并在其上删除了一些开源ML算法。 通过
云标准它是一个可笑的小数据集，但它应该处理得相当快
我注意到你只有一类无监督算法，但你有几个神经网络。 您的神经
网络可以处理未标记的数据以充当自动编码器或自组织映射吗？
Azure的ML算法是否针对您的HPC解决方案进行了优化？ 我知道我的数据集并不值得HPC时间，
但它很有趣。
可以将ML算法配置为相互馈送，特别是K-means可以作为神经网络的输入吗？ 在过去的几年中，有一些研究表明无监督算法对零日攻击的有效性，还有一些研究显示假阳性率为40％ -60％用于网络安全卷中的无监督算法。 我对$ 0 $ b怀疑零日检测，他们可能只是标记所有内容并且变得幸运。 我想使用
无监督算法的输出作为监督模型的输入要素的一部分，如果我错了，监督的模型应该清理误报，同时仍然是零日。
Azure的机器学习工作室能满足我的需求吗？
感谢您的时间，
Kieran
< blockquote>
我将您在机器学习服务中的帖子移到了这个论坛。根据我对您的问题的理解，Machine Learning Studio应该是一个不错的选择，我会建议您尝试使用Free trail查看。 https://studio.azureml.net/
您需要考虑的一点是，您的数据集有多大。如果您的数据集大于50GB，那么您需要考虑如何处理它。
总之，Azure机器学习工作室很适合您的上述要求，并且很容易为新用户启动。
有关Azure机器学习工作室的指导，请参阅 https://docs.microsoft.com/en -us / azure / machine-learning / studio / what-is-ml-studio
------- -------------------------------------------------- --------------------------------------------------
如果您发现此信息有用，请将其命名为"有用"。如果他们有帮助，请记得将回复标记为答案。
问候，
宇通
I asked this on the Azure Machine Learning service and way advised to redirect my question to the AML studio forum.
Hi,
I'm working on a Machine Learning (ML) experiment for my Master's thesis. I've been advised that my school might be able to support my renting some cloud computing time Azure looks very promising, but I have questions.
First an overview of my project. ML is a hot topic, in a lot of realms I'm looking at cyber security, specifically data exfiltration. I have a data set, about fifty gigabytes of packet capture (PCAP) data from a simulated network, processing this on a cloud solution would faster than building a desktop and dropping some open source ML algorithms on it. By cloud standards it is a laughable small dataset, but it should process fairly fast
I noticed that you only have one class of unsupervised algorithm, but you have several neural networks. Can your neural networks process unlabeled data to act as autoencoders or self-organizing maps?
Are Azure's ML algorithms optimized for use with your HPC solutions? I know my dataset doesn't really merit HPC time, but it'd be fun.
Can ML algorithms be configured to feed into each other, specifically can the K-means act as an input into say a neural network? There has been some research in the past couple of years touting unsupervised algorithm's effectiveness against Zero-day attacks, there is also research showing a false positive rate of 40%-60% for unsupervised algorithms in cyber security rolls. I’m skeptical about the zero-day detection, they might just be flagging everything and getting lucky. I’d like to use the output of an unsupervised algorithm as part of the input features of a supervised model, if I’m wrong the supervised model should clean up the false positives, while still catching zero-days.
Would Azure’s Machine Learning Studio meet my needs?
Thank you for your time,
Kieran
Hi,
I moved your post in Machine Learning Service to this forum. From my understanding of your question, Machine Learning Studio should be a good choice for you, I would suggest you try with the Free trail to see. https://studio.azureml.net/
One point for your consideration is, how big is your dataset. If your data set is larger than 50GB, then you need to think about how you want to deal with it.
As a conclusion, Azure machine learning studio is good for your requirement above, and it is very easy to start for new user.
For guidance of Azure machine learning studio, please see https://docs.microsoft.com/en-us/azure/machine-learning/studio/what-is-ml-studio
-----------------------------------------------------------------------------------------------------------
If you found this post helpful, please give it a "Helpful" vote.
Please remember to mark the replies as answers if they help.
Regards,
Yutong
这篇关于Azure机器学习功能问题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！
