Danilo J. Rezende
@danilojrezende.bsky.social
VP of AI Research, Principal Scientist @ EIT Oxford | ex-Director @ DeepMind Building models to accelerate fundamental sciences and medicine.
Opinions my own.
https://danilorezende.com/
Opinions my own.
https://danilorezende.com/
Amazing, congrats!
January 16, 2025 at 11:57 AM
Amazing, congrats!
Indeed, additionally If computing the forces in HMC is computationally cheap - it's extremely hard to beat it with learnt samplers in terms of wall clock time. Similar challenge to Neural PDE solvers.
December 27, 2024 at 11:33 PM
Indeed, additionally If computing the forces in HMC is computationally cheap - it's extremely hard to beat it with learnt samplers in terms of wall clock time. Similar challenge to Neural PDE solvers.
Should cite Newton first 😂
December 24, 2024 at 10:10 PM
Should cite Newton first 😂
Very nice results!
December 21, 2024 at 2:48 AM
Very nice results!
Amazing work!
December 7, 2024 at 1:38 AM
Amazing work!
Just use github
December 4, 2024 at 8:20 AM
Just use github
As a statistical inequality hoarder, I approve 😅
November 30, 2024 at 9:15 PM
As a statistical inequality hoarder, I approve 😅
What do you mean by physical? Eg If the target density lives on a specific manifold, RFs will generally not respect that naively (eg think of data living on a torus). One would have to choose paths along geodesics for example and then properly integrate the flow on the manifold.
November 30, 2024 at 4:03 PM
What do you mean by physical? Eg If the target density lives on a specific manifold, RFs will generally not respect that naively (eg think of data living on a torus). One would have to choose paths along geodesics for example and then properly integrate the flow on the manifold.
Yes, though a simplification is that the log-det-jac is the integral of the divergence. So "just" need to solve a 1d integral.
November 30, 2024 at 10:03 AM
Yes, though a simplification is that the log-det-jac is the integral of the divergence. So "just" need to solve a 1d integral.
You can, just need to integrate it's divergence to get the likelihood.
November 30, 2024 at 10:01 AM
You can, just need to integrate it's divergence to get the likelihood.
What do you mean? That's the whole field of optimisation (preconditioning, proximal, natural, etc...)
November 29, 2024 at 8:22 AM
What do you mean? That's the whole field of optimisation (preconditioning, proximal, natural, etc...)
I have quite a few of those 😅
November 27, 2024 at 8:36 AM
I have quite a few of those 😅
Reposted by Danilo J. Rezende
For the theoretical side, there’s no better resource than the 2024 book by Blondel and Roulet:
arxiv.org/abs/2403.14606
For the practical side, my friend @willtebbutt.bsky.social wrote stellar documentation explaining autodiff with mutation:
compintell.github.io/Mooncake.jl/...
arxiv.org/abs/2403.14606
For the practical side, my friend @willtebbutt.bsky.social wrote stellar documentation explaining autodiff with mutation:
compintell.github.io/Mooncake.jl/...
The Elements of Differentiable Programming
Artificial intelligence has recently experienced remarkable advances, fueled by large models, vast datasets, accelerated hardware, and, last but not least, the transformative power of differentiable p...
arxiv.org
November 26, 2024 at 6:05 AM
For the theoretical side, there’s no better resource than the 2024 book by Blondel and Roulet:
arxiv.org/abs/2403.14606
For the practical side, my friend @willtebbutt.bsky.social wrote stellar documentation explaining autodiff with mutation:
compintell.github.io/Mooncake.jl/...
arxiv.org/abs/2403.14606
For the practical side, my friend @willtebbutt.bsky.social wrote stellar documentation explaining autodiff with mutation:
compintell.github.io/Mooncake.jl/...