Re: [Nmlrg] Generic knowledge definition and format?

[Albert Cabellos <[e-mail]>]

Hi all

We have two experiments demonstrating use-case1 (Overlay Routing) and
use-case 2 (VNF):

Overlay Routing: We can learn how to route in an overlay network when
the underlay is hidden, this means that we learn the function that
relates the traffic matrix on the overlay, routing policy on the
overlay (traffic to paths defined as egress and ingress interfaces of
the overlay nodes) with the underlay performance.

VNF: We can learn the cost (CPU cycles) of a hidden VNF (black box) as
a function of the traffic that it is processing, in this case the
traffic is defined as 70 traffic features.

Unfortunately we don´t have yet an experiment for knowledge
extraction, mainly due to the lack of real-world data. Knowledge
extraction should (in my view) use statistical techniques such as
K-means, K-nearest neighbors and Correlation Analysis. Such techniques
operate over the data-set and hence, the result have the same
semantics than the data-set.

Albert

On Tue, Nov 17, 2015 at 9:04 AM, Sheng Jiang <[e-mail]> wrote:
> Hi, Albert,
>
> Thanks for your interesting presentation in the proposed NMLRG meeting in Yokohama. It is a pity we do not have enough time discuss. However, mail list discussion may give us more flexibility to reach deep details. Since you have multiple use cases, let's discuss one by one for better communication.
>
> In your use case 4: knowledge extraction, have you already done some prototype? Would you share some concrete example of knowledge? Actually, I am thinking whether it is feasible to give a generic definition of knowledge and whether it is possible to design a generic format/language to describe knowledge. If so, these knowledge may be much easier to be understood and trigger follow up actions.
>
> Regards,
>
> Sheng