Weekly report, week of Jan 13
(sorry this is late, but I've been out of commision for the past few days
with intestinal flu - and very nasty it was, too!)

NHSE
----

Talked to David about the chemistry repository, and we are registering csir.org
domain and putting out an ad looking for a chemistry student to work part-time
on the repository. We also discussed created repositories for the DoD
modernization program, and he is going to try to identify people who would be
willing to work with us on that.

I've been having a lot of trouble getting the RIB installed here, but I think
I've finally sorted out the problems and got the help and information I needed 
from systems, so I'm hoping everything should be up and running in the next 
couple of days.

CSNet
-----

Software for Gary is coming along fairly well. He is hoping that his MCI 
deal will roll out in about 8 weeks, and thus wants a solid version of the 
software ready in about 4 weeks, so his people can test it, help write 
documentation, etc. I think that should be feasible without too many late
nights.


Other stuff
-----------

I installed Windows 95 version of Tango on my laptop. In order to give a
sensible demo, another machine is needed to demonstrate the collaborative
capabilities. When I get the new PC at the Lubin House I'll install it there
so I can demo it properly over 2 machines.

Finally cracked the one nagging problem that's been bothering me about Saleh's
work and papers. Some time ago he was able to achieve the goal of scheduling
all SU classes using his annealing algorithm, but then he realized that he'd
misinterpreted some of the class data that SU gave him, and in fact only about
95% of the classes were scheduled. I suggested some improvements to the
algorithm, and some parameters to tweak, and we got it up to about 99%, but
that was still very annoying, because we couldn't claim to have truly solved
the problem. This week I finally realized that we were being too hard on 
ourselves, in that the hard constraints included not just the scheduling of 
the classes, but also the hard student constraints, e.g. that a student cannot
be in more than one class at the same time. But these should actually be
"medium" constraints, i.e. we should try to satisfy them as much as possible, 
but if some students' schedules don't work out, then they should just be told 
that their preferences cannot be satisfied and they have to change them 
(which is of course what is done in the semi-automated process currently 
used by SU). In other words, the scheduling of the classes should take higher 
priority than satisfying student preferences. The problem was that Saleh's code
was giving these roughly equal priority, which was making the problem virtually
impossible to solve. He has now changed this, and the program is now producing
a valid schedule for all the classes every time, and usually satisfying better 
than 90% of student preferences. He is currently getting new results and plots 
using the new algorithm, and I am doing some minor editing of the paper, and 
then it will finally be finished. He is also writing a version for submission 
to a scheduling conference in Toronto this year. Once these are done, the plan 
is to finish off the paper on using mean-field annealing (basically saying that
it doesn't work), and then to get him started writing up his thesis.