Re: [Nmlrg] Machine Learning in network - solicitation for use cases

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

Hi Sheng,

> -----Original Message-----
> From: Sheng Jiang
> Sent: Saturday, September 19, 2015 10:12 AM
> To: Liubing (Leo); Sebastian Abt
> Cc: Brian E Carpenter; Dacheng Zhang; [e-mail]
> Subject: RE: [Nmlrg] Machine Learning in network - solicitation for use cases
> 
> >[Bing] Indeed. The trick/art is in feature selection.
> 
> Not only the feature selection, the learning direction or path design are also
> important. It needs the implementors/designers to apply the specific prior
> knowledge to indicate/guide the mechanism learning process. The good
> design with valuable prior knowledge would enhance the efficiency and
> accuracy of the machine learning application. However, the more prior
> knowledge applied, the less generality it would be.

[Bing] For " learning direction or path design ", did you mean this: one application could be divided into multiple parts or stages, each part/stage might involve different learning models/algorithms (or maybe the same learning models/algorithm but for different features at each stage). Then the "learning direction or path" is about how to separate the stages and choose what models/algorithms. It indeed needs more human wise involved. And I guess maybe it is more practical in real application?

B.R.
Bing

> Sheng