I met with Dr. Abraham today to discuss my
Grid Security paper, and his ideas to include statistical trust models into it. I managed to convince him that the distributed model of trust would be better. The argument went something like:
Me If I use a Bayesian model, then each node would have to store historical information about interactions with every other node, correct?AA:CorrectMe: Then I'm storing a magnitude of information size N on each node, that's a lot of space. I can accomplish effectively the same goal using referrals and I would only have to store a fraction of the history information required of a Bayesian model.AA:Yes, I see. The idea is sound.
I went to on to explain the ways in which we could extend this idea, mainly with the use of Ontologies to describe the Applications language (or protocol if you will). You could then use this as a baseline to detect system anomalies. Two hours later, I leave his office with a stack of (mostly) crap papers, and two rejected NSF grant proposals to read. Meanwhile, that time would have been better spent working out my DS2 homework: dynamic programming solution to the activity scheduling problem.
I saw the Matrix Revolutions tonight, and after a long day of battling with dynamic programming it was exactly what I needed. After dropping Maryna back at her apartment I started the drive home and realized my mistake: I was trying to use a matrix like the book suggest doing, but my implementation is far beyond using a plain table. The approach for dynamic programming is to calculate the answers to the subproblems and save them so that they can be looked up when you need them. Storing sets, like I am using, in tables isn't practical - I need to be saving the subsets in a structure which can be referenced more easily than a table.
