[This note is directed mostly at the students in my graduate architecture class, but am copying the nlp at umd list as it might be of interest there too...]
It was good to see so many of you at William Lokke's talk - it's fair to say he's done well since graduating from UMD in 1957 - in this talk he was in effect trying to summarize a 40 year career at Lawrence Livermore Labs so he couldn't get into any one topic in depth, but indeed there was some discussion of supercomputing as we expected. There was an informal gathering with him before the talk that I attended, and he actually talked there with great passion about supercomputing, not only how important it was at LLNL, but also the pressing need to solve the multi-core problem, and how programming is going to change forever and we need to do something about it. By we I think he meant CS programs, so I suppose we are taking small steps in that direction with our adventures in MPI (one hopes?). What was also interesting in that informal meeting is that he is quite interested in Artificial Intelligence. He writes neural network programs in his spare time, thinks about brain imagining experiments a lot, has written checkers and chess programs, and thinks about the issues of consciousness (what does it mean to be conscious, and other great questions of philosophy). I found all this a bit hard to recognize in the first part of his talk (which was mostly about nuclear weapons), but then of course when he got to the Grand Challenge problem he posed at the end of the talk, I immediately understood. The Human Brain is indeed the greatest of grand challenge problems not only in AI but probably in all of science, there is no bigger one, and no harder one, and I think the idea of creating a mechanical human brain is really at the core of what gets a lot of people into computing in the first place, or what they inevitably start to think about once they have been around computers for a while. I was not familiar with the project he mentioned, the Blue Brain Project, but I found their web page and it's pretty interesting. http://bluebrain.epfl.ch/ IBM is the sponsor of this, and they are using supercomputers and parallelism (as the brain is amazingly parallel). It seems they are using one of IBM's Blue Gene systems for this work, so this is a very big time operation I think (or if nothing else they are throwing a lot of computing power at the problem). Just to record some of his figures on the brain.... There are 100 billion neurons, 100 trillion synapses, and 100 neural firing per second in the human brain So, If you think of the neurons as nodes, the synapses as links between the nodes, and the firings as messages being passed.....well, you see why people in supercomputing sometime get excited about the human brain (and vice versa). Interestingly enough, the way I have heard it described (and what these numbers support) is that the human brain is incredibly parallel, but not incredibly fast (as the 100 firings per second suggest, if you equate that with clock speeds on CPUs it's terribly slow...) This is actually sort of what people say about multi-core computers - more parallel, not as fast...so the metaphors with the human brain become even stronger these days I think than they have been in the past... Finally, I very much like the quote he attributed to E.O. Lawrence (founder of LLNL) : "No failures means you're not being bold enough." I haven't heard that said exactly that way, but I rather like it. I think it's true, just don't fall into the trap of thinking that everything you do is bold. :) In any case, glad so many of you were able to make it and I hope you found it thought provoking in some way. Ted --- Ted Pedersen http://www.d.umn.edu/~tpederse
