Autonomy and Perception; current @anduriltech, early and Director of Eng at @RebellionDef. AI/ML, SRE, Lawyer. Former General Counsel and engineer @USDS.
@thosjleeper Neal Stephenson's Anathem, which is about a thousand pages. I bought it used for a great price too, so it's also the best dollar per page ratio of any book I own.
@theshawwn@gwern I have had good results normalizing around the mean/std of the specific training dataset during GAN training. I have specifically tested this and ended up with better overall results versus doing something like normalizing around -1/1. Worth a try!