Subject: Re: CSIT curriculum Resent-Date: Thu, 06 Jan 2000 07:25:44 -0500 Resent-From: Geoffrey Fox Resent-To: p_gcf@npac.syr.edu Date: Tue, 28 Dec 1999 11:58:24 -0500 From: "Bernd A. Berg" Reply-To: "Bernd A. Berg" To: "Geoffrey Fox" CC: "Larry Dennis" Geoffrey, >So we need to have a plan that covers your material; useful stuff from my > older lectures; what Mascagni in CS can do -- this would be multiple > courses but we should think how they fit together and if you need to add > or subtract basic topics Clearly, you thought more than anyone here about curricula questions. Therefore, I think, you should take the lead and I am certainly very willing to help in my limited area of expertise. My script emerged from teaching an introduction to Monte Carlo (MC) Simulations and Statistics to physics graduate students at FSU, in Vienna and at a lattice gauge theory school in Trento. There, the focus was to get the students, who mainly specialized on either statistical physics or lattice gauge theory, quickly towards running simulations, but without neglecting the fundamentals. For the CSIT curriculum we are facing a somewhat different situation, because we are in the fortunate situation that we can build up a broad education. Also, we do not want to specialize on physics students. To have something to start with, I pulled the table of contents of my script out. I started working on it only some while after my 97 Trento lectures and try to broaden the target group. The course includes Fortran code (C versions planned). My statistics teaching is novel in the sense that I first introduce (Marsaglia) uniform random numbers as "god" given and subsequenly illustrate all concepts by simulations. Students become used to MC long time before importance sampling is introduced. Let me address the following questions: (a) Which material should be added to mine? (b) Should the material stay together in one or two courses or become divided up and integrated in other courses? (c) What material should become an integral part of the CSIT core courses (required for all CSIT graduate students) and what should be moved to more specialized advanced courses? Concerning (b) and (c): ================== My opinion is that there should be one 3h credit course in the CSIT core part with the title 1. "Stochastic Simulations and Statistics" (or similar), and my understanding is that there will be another 3h core course to cover 2. PDEs. This would still leave two (or more) 3h core courses (presumably to be taught first) for 3.+4.+... "Introduction to Information Technology" and "Computational Science" to the extent not covered by 1.+2. Integrating the material of 1.+2. into general courses will most likely water them out. Stochastic Simulations and PDEs are core methods and it is by methods and enabling tools that diverse multidisciplinary interests can find common ground. Students deserve that experts on Stochastic Simulations or PDEs will teach those parts. For instance, I would not consider myself particularly qualified (or interested) to teach about PDEs, but would be interested in picking up course 1. On the other hand, to integrate MC simulations into the general courses and let PDEs stand out, creates a serious imbalance of methods. Stochastic simulations are central for many applications in physics, chemistry and structural biology, but some CSIT related faculty appears to be be uneducated about it. To have the wrong people teach stochastic methods could seriously harm multidisciplinary integration. Concerning (a): ============= I scanned through your old courses for additional material, may be Mascagni has more to add. Before finalizing things we need to decide on the course structure. On the basis of my opinion on (b) and (c) one may break up the material into a core course and and advanced topics specialized course. In the order your links I find the following additional material: Course I (your statistics for physicists course): ======= Borel sets - axioms, Bayes Theorem and applications, Special distributions (Binomial, Bernoulli, Poisson), Fitting (estimation of parameters): Maximum likelihood, (Least square) Minimization and Marquardt's method. Remarks: These topics were actually included in the Special Topics course I taught at FSU, but they did not yet make it into my script. I introduce goodness of fit early with the Gaussian difference test. Jackknife ought to be taught next and fitting later, but details are finally up to the individual teacher. Also: Difference tests for distributions (chi2 and Kolmogorov); maximum entropy methods (?) deserve to be included. Further: You give many examples, mainly from physics. For the curriculum a statement like "examples from basic and applied sciences" is presumably sufficient and allows to broaden into chemistry and structural biology. Course II (your MC simulations for stat. phys. course) ======= Multigrid, Simulated Annealing, possibly hybrid MC (leapfrog) should be added to the CSIT curriculum. The rest is either already covered by my table of contents or would fall under "examples from basic sciences". Course III (your numerical integration) ======= MC for multidimensional integrals should be added. The other MC parts of your course are already covered and the non-MC parts should be scanned for inclusion in a CSIT PDE course or other CSIT courses. Course IV (Your statistics and random numbers) ======== The topics are, in principle, already included. Please let me know how you think we should proceed. Happy y2k! Bernd