iiinova — where your digital self keeps evolving

Andrei Manolache

@amanolache.bsky.social

255 followers 186 following 11 posts

ELLIS PhD Student @ IMPRS-IS/Uni Stuttgart; ML Research @ Bitdefender | andreimano.github.io

Posts Replies Media

Reposted by Andrei Manolache

EEML @eemlcommunity.bsky.social · 15/12/2024

EEML'25, our yearly machine learning summer school event, will be organised next summer in the beautiful city of Sarajevo - the place where East meets West 🇧🇦🇧🇦🇧🇦. More details coming soon, please see the link in the thread!

1196

Andrei Manolache @amanolache.bsky.social · 15/12/2024

Catch my poster tomorrow at the NeurIPS MLSB Workshop! We present a simple (yet effective 😁) multimodal Transformer for molecules, supporting multiple 3D conformations & showing promise for transfer learning. Interested in molecular representation learning? Let’s chat 👋!

0112

Andrei Manolache @amanolache.bsky.social · 11/12/2024

* in the afternoon session 😅

000

Andrei Manolache @amanolache.bsky.social · 11/12/2024

Happening today, East Exhibit Hall A-C, poster #3110. Come say "Hi!"! 👋

110

Andrei Manolache @amanolache.bsky.social · 07/12/2024

6/6 Interested in learning more? Check out our preprint here: arxiv.org/pdf/2405.17311. If you’d like to discuss, I’d be very happy to chat during the poster session in Vancouver! :)

arxiv.org

210

Andrei Manolache @amanolache.bsky.social · 07/12/2024

5/6 How it works: Probabilistic sampling connects original nodes to virtual ones, enhancing connectivity without explicit pairwise computations. The result is a framework that achieves both higher WL expressiveness and efficiency in graph-based learning.

100

Andrei Manolache @amanolache.bsky.social · 07/12/2024

4/6 We demonstrate SOTA results on various benchmarks, effectively addressing over-squashing and under-reaching. IPR-MPNNs also surpass standard MPNNs in expressiveness, distinguishing complex graph structures—all while being faster and more memory-efficient than GTs. 🚀

100

Andrei Manolache @amanolache.bsky.social · 07/12/2024

3/6 Enter IPR-MPNNs: Our approach learns to rewire graphs probabilistically by adding virtual nodes. This eliminates the need for heuristics, making the method more flexible and task-adaptive, while maintaining computational efficiency. 🎯

110

Andrei Manolache @amanolache.bsky.social · 07/12/2024

2/6 Standard MPNNs struggle with long-range interactions, making them less effective for large, complex graphs. Transformers help but come with quadratic complexity, which is computationally expensive. Rewiring heuristics? Often brittle and task-specific.

100

Andrei Manolache @amanolache.bsky.social · 07/12/2024

1/6 We're excited to share our #NeurIPS2024 paper: Probabilistic Graph Rewiring via Virtual Nodes! It addresses key challenges in GNNs, such as over-squashing and under-reaching, while reducing reliance on heuristic rewiring. w/ Chendi Qian, @christophermorris.bsky.social @mniepert.bsky.social 🧵

1306

Andrei Manolache @amanolache.bsky.social · 29/11/2024

Genuine question - why do captions go above tables? I've always assumed that this is due to wanting to make tables distinct from figures, but it seems like a convention.

120

Andrei Manolache @amanolache.bsky.social · 18/11/2024

👀👋

110