iiinova — where your digital self keeps evolving

Reposted by Sonia Mazelet

NeurIPS Europe @neuripseurope.bsky.social · 28/08/2026

The first call for NeurIPS Paris 2026 volunteers is open! We are seeking for PhD students and postdocs to assit during the conference in exchange for complimentary registration. Application form: forms.gle/HGDUzS7y7yvq... 1st call deadline: September 10th

01610

Sonia Mazelet @soniamazelet.bsky.social · 05/07/2026

I’ll be at ICML presenting our paper! Come say hi and check out our poster on Thursday 9th (Hall A, poster #3710). Huge thanks to @simonroschmann.bsky.social @paulkrz.bsky.social, @qbouniot.bsky.social and @zeynepakata.bsky.social for this great collaboration !

051

Sonia Mazelet @soniamazelet.bsky.social · 26/11/2025

I am happy to share that I will be at NeurIPS in San Diego to present our paper with @rflamary.bsky.social and Bertrand Thirion on optimal transport plan prediction between graphs. If you are around come say hi! Paper: arxiv.org/abs/2506.12025 Poster #3703 Friday 5 December 4:30 - 7:30 pm

arxiv.org

Unsupervised Learning for Optimal Transport plan prediction between unbalanced graphs

Optimal transport between graphs, based on Gromov-Wasserstein and other extensions, is a powerful tool for comparing and aligning graph structures. However, solving the associated non-convex optimizat...

0143

Reposted by Sonia Mazelet

Paul Krzakala @paulkrz.bsky.social · 16/10/2025

Our latest paper “The Quest for the GRAph Level autoEncoder (GRALE)” was accepted at NeurIPS 2025! arxiv.org/abs/2505.22109 🏆 GRALE 🏆 can encode and decode graphs into and from a shared Euclidean space. Training such a model should require solving the graph matching problem but...

arxiv.org

The quest for the GRAph Level autoEncoder (GRALE)

Although graph-based learning has attracted a lot of attention, graph representation learning is still a challenging task whose resolution may impact key application fields such as chemistry or biolog...

172

Sonia Mazelet @soniamazelet.bsky.social · 29/09/2025

We apply ULOT to the problem of brain alignment and find that it predicts near-optimal FUGW plans up to 100× times faster than other solvers. This efficiency enables detailed exploration of the effects of the FUGW hyperparameters on the optimal plans, and many more applications! (5/5)

080

Sonia Mazelet @soniamazelet.bsky.social · 29/09/2025

ULOT predicts FUGW plans conditioned on the FUGW hyperparameters. Trained in a fully unsupervised way by minimizing the FUGW loss, it ensures the near optimality of its predictions and diminishes the complexity of finding optimal FUGW plans from cubic to quadratic in the number of nodes. (4/5)

130

Sonia Mazelet @soniamazelet.bsky.social · 29/09/2025

Matching graphs can be achieved with the optimal transport distance Fused Unbalanced Gromov Wasserstein (FUGW). It produces meaningful plans but requires solving an optimization problem with cubic complexity in the number of nodes, which limits its applications. (3/5)

130

Sonia Mazelet @soniamazelet.bsky.social · 29/09/2025

We developed ULOT, a neural network designed to predict optimal transport plans between graphs. It achieves accurate predictions up to 100× faster than solvers, both on synthetic graphs and on brain data. (2/5)

130