Simon Knowles on pushing AI computing to the limit by rethinking chips
The Robot Brains Podcast
00:00
Signal Processing - Layers Are Like the Components That We Build
We find that the best way to learn things s is incrementally eloa b by many layered transformations of relatively small effect. We're just learning what sort of set is useful, and in future, once we've got a good set of parts, we'll just bolt them together in various ways. I'm quite fascinated at the moment by the convergence of traditional see and end and a attention type models might transform us. And it's wonderful to see that good results are being achieved by using a few san en layers at the front and then a few transformer layers at the back.
Transcript
Play full episode
