
Episode 16: Yilun Du, MIT, on energy-based models, implicit functions, and modularity
Generally Intelligent
00:00
The Attractor Dynamic Theory of Memory
Energy base models have some connection as well with the brain, as the bram love's not wasting too energy. Theres a lot of like, analogues you see in a few different places to this energy minimization that's acually quite interesting. And if you think about the normalizing factor makes no sense for people like we don't have calebrated probabilities, but we certainly know if something is more or less likely. Or i seems much more analogous to the types of things we do, these energy functions and probabilities. A raw iremember would be even more powerful than usa. Unfortunately, at least we don't hae agoo og of tem for training them
Transcript
Play full episode