control problems in communication networks

  • To: ripe-list@localhost
  • From: RIPE NCC Staff ncc@localhost
  • Date: Thu, 19 Aug 1993 16:52:22 +0200
  • Cc: bleich@localhost

Can anyone assist Mario with his query below?

------- Forwarded Message

Date:     Thu, 19 Aug 93 16:29:40 +0200
From:     bleich@localhost
To:       ncc@localhost
Subject:  control problems in communication networks

Dear Anne,

let me explain what I want. In our lab we deal with machine learning algorithms
,
i.e. decision tree algorithms.
The generic problem to machine learning is as follows. From samples of data 
(exapmles) described by a vector of attributes and a class, derive a function
which predicts the class of the samples from the same population. The general
scheme is to present the algorithm with "training data", from which it "learns"
a classification function, then apply this function to some unseen "test data"
to predict their class.

One example for this is the recognition of handwritten digits. The handwritten
digits were digitised onto images with 16*16 pixels and 256 grey levels.
So each example consists of 16*16=256 attributes and 1 class (the digit).
The resulting decision tree does not need all attributes to predict the class,
only those which are relevant for the classification.

Other application problems are in medicine, banking, etc.

The question is now, is such an algorithm applicable on problems in
communication engineering? One of the problems for example is the routing and 
flow control in networks. Perhaps it is possible to simulate such control
algorithms. Mr. Kaehler told me, that you have made some experiments with
the IBORN. It would be nice, to hear more about that. May be, you have
data to another classification problem (like error recognition), too.

Regards
Mario Bleich

------- End of Forwarded Message