Looks like we don't have to change too much. Overall the reviews are very
good.

The first and third reviews wanted a bit more information on the expert
system. We could probably add that using some of the new stuff on the ES
that was added in the conference paper. I could do that and you can review
the changes.

First review also wants some more info about MFA. You should do that and
show me your changes so I can review them.

Also from 3rd review:
 - Don't we say that ES results were not averaged over 10 runs because it
   is deterministic? If not, need to add.
 - For SA, do runs using random initial state start at higher temp than runs
   using initial state from ES? If not, why not? If so, how is the initial
   temp determined for the random starts?
 - I thought sparseness ratio was defined in the paper? If not, need to add.
You should check these points.

Second review is rather bogus and basically contradicts the results in the 
paper! There is only no reason to believe a good starting point won't help
if you have a perfect cooling schedule and a perfect move strategy that 
explores the search space extremely efficiently - in practice you have 
neither. I might add something to this effect.

You should get copies of the 2 papers mentioned in reviews 2 and 3 and I 
will read them and finalize changes to the conference paper, and hopefully
finish final version of the journal paper, when I am next in Syracuse - 
probably the week of the 12th or the week of the 19th.