Yannic Kilcher Videos (Audio Only) cover image

Neural Networks are Decision Trees (w/ Alexander Mattick)

Yannic Kilcher Videos (Audio Only)

00:00

GPUs Aren't Suited to GPUs

If one were to implement the decision tree on like a GPU one would actually regain all of these multiplications and additions because it just makes more sense to put the binary vector there with a lot of zeros. Decision trees really like to have these very very well-defined statistics because that helps them to do their splits effectively neural networks scale with gradients so if you can't get gradients you have a hard time  and they also scale with size simply because as we've seen here you just get more possible more representational power.

Transcript
Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
App store bannerPlay store banner
Get the app