Durk Kingma
@dpkingma.bsky.social
Research scientist at Anthropic. Prev. Google Brain/DeepMind, founding team OpenAI. Computer scientist; inventor of the VAE, Adam optimizer, and other methods. ML PhD. Website: dpkingma.com
Actually, my bad, there's code: github.com/sinhasam/CRVAE
GitHub - sinhasam/CRVAE: Official code release for Consistency Regularization for VAEs
Official code release for Consistency Regularization for VAEs - sinhasam/CRVAE
github.com
December 4, 2024 at 11:57 AM
Actually, my bad, there's code: github.com/sinhasam/CRVAE
Impressive!!
I'm curious if anyone tried to reproduce that 2.51 number from CR-NVAE. It's much lower than I would suspect for a method like hat so I'm curious if it's real, afaik they didn't share code.
I'm curious if anyone tried to reproduce that 2.51 number from CR-NVAE. It's much lower than I would suspect for a method like hat so I'm curious if it's real, afaik they didn't share code.
December 4, 2024 at 11:41 AM
Impressive!!
I'm curious if anyone tried to reproduce that 2.51 number from CR-NVAE. It's much lower than I would suspect for a method like hat so I'm curious if it's real, afaik they didn't share code.
I'm curious if anyone tried to reproduce that 2.51 number from CR-NVAE. It's much lower than I would suspect for a method like hat so I'm curious if it's real, afaik they didn't share code.
Probably my biggest naming blunder was the name "auto-encoding variational Bayes" for a method that (in its default version) is only Bayesian over the latent variables, not the parameters. Blasphemy in the church it Bayes 😂
December 3, 2024 at 1:14 PM
Probably my biggest naming blunder was the name "auto-encoding variational Bayes" for a method that (in its default version) is only Bayesian over the latent variables, not the parameters. Blasphemy in the church it Bayes 😂
Yeah, but I was born in darkness. The first paper I ever cited was Aapo Hyvarinen's paper introducing score matching from 2007, in which the term is "abused" that way the first time, but he had the excuse since there were no good alternatives! Inference, however...
December 3, 2024 at 1:11 PM
Yeah, but I was born in darkness. The first paper I ever cited was Aapo Hyvarinen's paper introducing score matching from 2007, in which the term is "abused" that way the first time, but he had the excuse since there were no good alternatives! Inference, however...
I completely agree that it's not an ideal situation that the meaning of the word inference is now overloaded, but its use in this context is now extremely widespread. Better embrace this strange new wor(l)d ;)
December 3, 2024 at 9:00 AM
I completely agree that it's not an ideal situation that the meaning of the word inference is now overloaded, but its use in this context is now extremely widespread. Better embrace this strange new wor(l)d ;)
My take is that the similarity stems from (1) the true score is identical across models except for stretching of time & space, (2) the only fundamental difference between diffusion objectives is the weighting, and (3) many common weightings are fairly similar (support over a similar range of SNRs).
December 1, 2024 at 10:18 AM
My take is that the similarity stems from (1) the true score is identical across models except for stretching of time & space, (2) the only fundamental difference between diffusion objectives is the weighting, and (3) many common weightings are fairly similar (support over a similar range of SNRs).
Do you mean the intuition behind the fact that flow matching with the optimal transport (FM-OT) objective = diffusion objective with exponential weighting? I should probably read the other paper you linked ;)
November 30, 2024 at 2:02 PM
Do you mean the intuition behind the fact that flow matching with the optimal transport (FM-OT) objective = diffusion objective with exponential weighting? I should probably read the other paper you linked ;)
No, CS. But intrigued by starlink's phased array :-)
November 27, 2024 at 7:26 PM
No, CS. But intrigued by starlink's phased array :-)