Jan-Willem van de Meent
banner
jwvdm.bsky.social
Jan-Willem van de Meent
@jwvdm.bsky.social
Associate Professor (UHD) at the University of Amsterdam. Probabilistic methods, deep learning, and their applications in science in engineering.
Reposted by Jan-Willem van de Meent
🐳 Controlled Generation with Equivariant Variational Flow Matching

By @eijkelboomfloor.bsky.social , @zmheiko.bsky.social ,
@sharvaree.bsky.social , @erikjbekkers.bsky.social ,
@wellingmax.bsky.social , @canaesseth.bsky.social *, @jwvdm.bsky.social *

📜 As above - paper TBA soon 😉

🧵8 / 8
May 6, 2025 at 3:03 PM
Reposted by Jan-Willem van de Meent
🌊 Exponential Family Variational Flow Matching for Tabular Data Generation

By Andrés Guzmán-Cordero*, @eijkelboomfloor.bsky.social *,
@jwvdm.bsky.social

📜 This paper will be shared soon - keep your eyes open! 🤩

🧵7 / 8
May 6, 2025 at 3:03 PM
Reposted by Jan-Willem van de Meent
🐸 Erwin: A Tree-based Hierarchical Transformer for Large-scale Physical Systems

By @maxxxzdn.bsky.social , @jwvdm.bsky.social , @wellingmax.bsky.social

📜 arxiv.org/abs/2502.17019

🧵2 / 8
May 6, 2025 at 2:53 PM
Nice work! We were ourselves also wondering why few approaches seem to consider conditional entropy as a function of time (beyond the time-warping in the work of @sedielem.bsky.social, as you note). Great to see that this can aid generation in other settings as well.
May 1, 2025 at 11:31 AM
I think that a key difference here may be that RPG mechanics are generally less stochastic than the conference reviewing process.
January 17, 2025 at 9:15 AM