During my visiting professor stint at Cornell, I subbed for Mitchell for one lecture - I told him that I could teach E&M without any formulas, only using birdtracks.eu notation, and I was lucky - got to teach kiddies Green's theorem. There might be someone out there who still wanders what happened that day.
I my clouds video I mentioned some brain related projects that didn't pan out. "Universality in chaos" AKA "period-doubling" does not care what particular system one starts with, and we were wondering how neurons - each one noisy and imperfect in every way - never-the-less build up a pretty robust brain. Could it be that the renormalized (?) composite neuronal system is more robust than all of its imperfect components? It was all very naive and totally wrong. That's not how computational neuroscience has developed since.
I actually spent that semester secretly (in an attic of the Math Department) coding my idea of how a passive observer (a frog on a rock) visually observes and then develops abstract concepts, i.e. learns. That bombed out in an spectacular way. I was using Barbie Burke's husband, John Hamal Hubbard's IBM workstation, and it 1) one day lost all the programs in its memory, and then 2) literally chewed up my back-up diskette when I tried to restore them, thus destroying without any trace (there were no printouts) months of work. That was the end of my secret foray into computational neuroscience. Looking today at ChatGTP billions of parameters, I was way too optimistic about how this would have worked.In other words, you do not want to get any neuroscience advice from Predrag .
No comments:
Post a Comment