Funabiki's neural network parallel algorithm

[na_victo]

Hi everyone.

I programmed the algorithm developed by Funabiki & Takefuji and
published in the IEEE paper "A
Neural Network Parallel Algorithm for Channel Assignment Problems in
Cellular Radio Networks".

I ran it with the first compatibility matrix and demand vector C1,D1
as inputs. Although the neural network converged to a feasible
solution, the convergence frequency wasn't 100% for this case as the
author said it would be.

After running the algorithm with real Philadelphia instances as input,
the network has never converged to a feasible solution, so I came to
the conclusion that I have misunderstood some part of the paper. A
professor at my university pointed out that there might be something
wrong in the paper, but gave no further assistance.

If anyone has ever implemented this algorithm or wants to help, please
do so.

na_victo.

[Gordon Stangler]

Well, to rule out issues with your implementation, start out with very
simple cases that you know converge. For instance, make your input
vectors identity vectors, and stuff like that. If everything checks
out, the issue may be either very subtle, or a problem with the paper.