[44] Hady Elsahar - NLG from Structured Knowledge Bases (& Controlling LMs)
The Thesis Review
How to Sample From Unnormalized Scorers
You could actually if you apply MCMC like MCMC on any score so what's that unnormalized distribution that satisfies your constraints is the score. Many people who work on EBM's will not call that an EBM but because it doesn't have a neural model inside but it don't have to learn an EBM you could also define an EBM. So then you've looked into both these I guess like in the original you would actually do this fine tuning procedure to match this distribution and then you said that recently you've been looking into just directly sampling from it.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
