NanoGuy

Deep Learning part 1

Back in the ol' days of the late 90s, starting around '96 I started a small software company named Ducer.net with a couple of high school buddies. You can find the last few years of our fading website in the internet archive wayback machine starting around 2001 but by then we had drifted far from our original purpose.

My original business plan had been built around neural nets and training them on distributed networked machines kinda like Folding@Home or Seti@Home etc. We would hornswaggle our friends and family into running a small screen-saver type app which would train neural nets and upload the resulting weight sets to a central server where they would compete in a genetic algorithm arrangement. Worked great but never earned us a penny or drew in any funding. The AI winter was in full force and no one would talk to an egghead with a big idea.

Now with the advent of GPUs and the new TPUs the sort of deep learning we were just starting to play with when we folded up in 2003 is beyond the "Hot New Thing" so I can't help but appreciate the subtle irony.

posted by Guy @ 6:48 PM