[13] Adji Bousso Dieng - Deep Probabilistic Graphical Modeling
The Thesis Review
00:00
How to Align AI With Human Digital Data
You can take any GaN loss and add this regularizer and use the same algorithm. And then here is this also kind of agnostic to the specific GaN loss that's used? Like, could you use this with the Wasserstein GaN? Yeah, yeah. And you can use it with different architectures. We looked at DC GaN, which was the first architecture that really worked worked well. Then we looked at more recent ones like Style GaN,. But yeah, it's also agostic to the underlying architecture or the original GaN loss. So maybe now we can talk about the future. So you'll be starting next year, next September as a professor at Princeton
Transcript
Play full episode
