Re: [Nmlrg] Using Machine Learning for Network Device Initial Configurations-//RE: Machine Learning in network - solicitation for use cases

["Liubing (Leo)" <[e-mail]>]

Hi Sheng,

> >We're working on a use case of applying Machine Learning technologies
> >to network device initial configuration.
> >The basic scenario is to learn knowledge/patterns of how to configure a
> >device from historical data (device configuration data of a number of
> >networks ) and then apply them to the new networks.
> 
> Hi, Bing,
> 
> Thanks for sharing. It would be helpful if you could introduce a little bit more
> on how to leverage the machine learning in your use case, such as the
> learning objectives, what a result of the learning, how it apply to a network,
> etc. 

[Bing] The program processes the database which is a collection of a number of existing/historical network information/configurations as the training data, to learn the correlation pattern between a specific parameter configuration and the network information. The network information includes network planning manifest, topology (mostly local partial topology around the device to configure), and every device's information including device manufacture, type, capacity etc.

The learning results are be kind of rules or functions that used for predicting configurations for a given device in a network to be configured. 
For example, the result could be a classification model. Because many of the parameter values could be considered as tags, even for some numeric value, it could also be divided into several value segments that are identical as tags.
When applying it to the new network, just input the device and network information, then we can get the output classification result ( the parameter value). 
Comparing to current device configuration approach, the input is towards minimum requirements for human operations. And it is very flexible, because the configurations are generated dynamically based on the input. 

There are various ways to perform machine learning in this use cases. The design space relies on the data pattern and algorithms chosen.

> It would also useful to discuss the precondition of your use case and the
> constrain of the machine learning mechanism, either in general way or
> specific in your use case.

[Bing] The most important precondition is the training data. The training data should contain a decent number of network samples. The minimum number depends on the quality of the data. And each network sample should contain complete topology information, network planning manifest and every device's information and complete configuration lines.

For constrains, in general there are two:
1) As you see, this is actually a Data-Mining process, which strongly consumes computing resources and time, especially when the training data is huge.
2) The results are only prediction, they are NOT guaranteed to be valid, and currently we lack a mechanism to evaluate them instantly.

B.R.
Bing


> Best regards,
> 
> Sheng
> 
> >This is to leverage the automation on network initial configuration and
> >only require a minimal set of input such as high-level network
> >planning, which includes architecture design, address pool/block
> >assignment, protocol selection etc.
> >
> >The advantages of using Machine Learning on network initial
> >configuration could possibly be:
> >- Saving human cost
> >- Flexibility
> >  - dynamically generating configurations on site
> >  - adaptive to different types of networks
> >- The ability to continuously optimize the parameters, if the program
> >also learning the running performance
> >- etc.
> >
> >Comments are welcomed.
> >
> >Best regards,
> >Bing
> >
> >
> >> -----Original Message-----
> >> From: nmlrg [mailto:[e-mail]] On Behalf Of Sheng Jiang
> >> Sent: Monday, August 31, 2015 11:16 AM
> >> To: [e-mail]
> >> Subject: [Nmlrg] Machine Learning in network - solicitation for use
> >> cases
> >>
> >> Hi, all,
> >>
> >> Thanks for subscribe to NMLRG (Network Machine Learning) mail list.
> >> As we know, there are already many ongoing researches for Machine
> >> Learning in network, in many areas. But up to now, there are few
> >> matured applications yet. So it is the time for a Research Group to
> >> work on this future-oriented technology.
> >>
> >> The first step would be to collect possible use cases: where the
> >> machine learning mechanism could be used in networks. The use cases
> >> does not need to be mature, but should have potential.
> >>
> >> Note that this topic is rapidly moving from academic research into
> >> practical application. Therefore, use cases from university
> >> environments, industrial research and development organizations are all
> welcome.
> >>
> >> Best regards,
> >>
> >> Sheng
> >> _______________________________________________
> >> nmlrg mailing list
> >> [e-mail]
> >> https://www.irtf.org/mailman/listinfo/nmlrg
> >
> >_______________________________________________
> >nmlrg mailing list
> >[e-mail]
> >https://www.irtf.org/mailman/listinfo/nmlrg