I’m pleased to share that I've been selected as a Fellow of the 2026 Thinking About Thinking Program!
I’m grateful to join a global, research-led ecosystem engaging thoughtfully with how AI is developed, governed, and understood.
@geometric-intel.bsky.social @ai-ucsb.bsky.social
Interested in the analysis of GNN expressive power, limitations of the WL test or more in general in the definition of complete metrics on graphs? 👀
Check out this super cool work led by
@martin.topology.rocks !🔥
We find it does great on non-isomorphic graph distinction by evaluating on BREC icml.cc/virtual/2024...
and as an inductive bias on ZINC (compared to other homomorphism and subgraph count methods)
Not all is lost because we can approximate our true metric by sampling computable graphs (of a nice treewidth) and also homomorphisms!
Under mild assumptions, we prove our metric:
1. Is 𝒸𝑜𝓂𝓅𝓁𝑒𝓉𝑒 regarding the isomorphic distinction.
2. Is 𝓅𝑒𝓇𝓂𝓊𝓉𝒶𝓉𝒾𝑜𝓃 𝒾𝓃𝓋𝒶𝓇𝒾𝒶𝓃𝓉.
3. Can 𝒹𝒾𝓈𝓉𝒾𝓃𝑔𝓊𝒾𝓈𝒽 induced subgraphs (even if a retraction exists)!
4. Has a 𝒻𝒾𝓃𝑒𝓇 𝓉𝑜𝓅𝑜𝓁𝑜𝑔𝓎 than the metric space induced by WL.
However... It's hard to compute :(
Inspired by the 𝘎𝘳𝘰𝘮𝘰𝘷-𝘏𝘢𝘶𝘴𝘴𝘥𝘰𝘳𝘧 distance, we define distortion as the maximum difference of the attributes of a graph 𝗚 and a graph 𝗚' through a homomorphism. We construct a metric from this 𝘥𝘪𝘴𝘵𝘰𝘳𝘵𝘪𝘰𝘯 by taking the homomorphism providing the infimum result and looking at both directions!
To kick off the PhD journey with @pseudomanifold.topology.rocks:
What are the limitations of the WL metric, and what is an 𝘪𝘯𝘧𝘰𝘳𝘮𝘢𝘵𝘪𝘷𝘦 𝘮𝘦𝘵𝘳𝘪𝘤?
We answer these questions with our 𝗚𝗿𝗮𝗽𝗵 𝗛𝗼𝗺𝗼𝗺𝗼𝗿𝗽𝗵𝗶𝘀𝗺 𝗗𝗶𝘀𝘁𝗼𝗿𝘁𝗶𝗼𝗻
arxiv.org/abs/2511.03068
@olgatticus.bsky.social, Kavir and @erikjbekkers.bsky.social
Two new videos on #topology 🍩 and #DeepLearning 🤖 just dropped:
- Part 1: youtu.be/p1GKAV0DAPI
- Part 2: youtu.be/_b19lhGRJmg
Both are part of my invited talk at the "OneMath World School 2026." Thanks for having me!
🚀 LOGML26— Mentor applications!
📍 Imperial College London | 13–17 July 2026
Mentor a project team. Attend talks, socials, networking events. 💸 Travel support available.
Deadline: 18 March 2026(AoE)
Apply: logml.ai/apply.html
#LOGML #MachineLearning #SummerSchool
All these insights resulted in a 9.5/10 thesis grade (whatever that means), which I hope serves as a hook to give it a read!
📖 On arxiv: arxiv.org/abs/2511.03068
Under mild assumptions, we obtain a metric and we show 𝘤𝘰𝘮𝘱𝘭𝘦𝘵𝘯𝘦𝘴𝘴. However, as all graphs enthusiasts might notice, this implies the GI problem.
That's not the end of the story because we show that we can sample graphs and achieve completeness in expectation :0 !
My thesis project embarks on this question, under the supervision of both @erikjbekkers.bsky.social , @pseudomanifold.topology.rocks and @olgatticus.bsky.social we devise a metric over graphs, 𝘵𝘩𝘳𝘰𝘶𝘨𝘩 the maps that exists between them. It also has a cool name!
𝐇𝐨𝐦𝐨𝐦𝐨𝐫𝐩𝐡𝐢𝐬𝐦 𝐃𝐢𝐬𝐭𝐨𝐫𝐭𝐢𝐨𝐧 (ⒽⒹ)
Two month flew by since the start of the PhD journey with @pseudomanifold.topology.rocks and now the time is right to share how our collaborations officially started, with a question:
How can we measure difference in (attributed) graphs beyond binary isomorphism testing ? 🌀
!!!
It may only be a band-aid, but we have just announced our new "Salon des Refusés" sessions for papers rejected due to space constraints: bsky.app/profile/euri...
Oh come on, a #NeurIPS rejection based on sensible, well-argumented reviews which show an understanding of the paper and its limitations?
HOW AM I SUPPOSED TO COMPLAIN NOW?
"Like all good stories, the tale of EurIPS begins in the kitchen…"
📣📣📣
We are proud to announce our #EurIPS Affinity Events which will take place during the main conference days.
Our Affinity Workshop Chairs have accepted three great proposals from well-renowned organizations
@wimlworkshop.bsky.social
Learning Theory Alliance
@queerinai.com
See you in Copenhagen!
📢 Interested in doing a PhD in generative models 🤖, AI4Science 🧬, Sampling 🧑🔬, and beyond? I am hiring PhD students at Imperial College London for the next application cycle.
🔗See the call below:
joeybose.github.io/phd-positions/
✨ And a light expression of interest: forms.gle/FpgTiuatz9ft...
Søren Hauberg (DTU) gives a quick tour of the site of EurIPS 2025: Bella Center, which features its own Metro stop, 13 min. from the airport, 9 min. to the center of town 🚊
The workshops focused on (in chronological order):
- Variational Inference (youtube.com/playlist?lis...)
- Optimal Transport (youtube.com/playlist?lis...)
- Parallel Computing (youtube.com/playlist?lis...), and
- Computational Physics (youtube.com/playlist?lis...).
#EurIPS sure is community-driven!
Across our Calls for Workshops and Affinity Workshops we received a whopping 50+ proposals! Our Workshop and Affinity Workshop Chairs are now busy reviewing and discussing all proposals, and we plan to have the decisions ready mid-September 🤞
We just wrapped up the 5th Summer Geometry Initiative (SGI), bringing 30 Fellows to our cohort of 150+ alumni, who go on to PhD/MS programs, industry jobs, and more. Despite overwhelming success, SGI26 is uncertain thanks to the changing funding landscape. Please help by helping identify sponsors!
Friday afternoon! Finally time to look back at a busy week, and ask oneself — "wait, what did I do, again?"
Let's push for the obvious solution: Dear @neuripsconf.bsky.social ! Allow authors to present accepted papers at EurIPS instead of NeurIPS rather than just additionally. Likely, at least 500 papers would move to Copenhagen, problem solved.
First promo photo appearance 🥹
Interested in #topological deep learning (TDL) but not sure where to start? 🍩🤖
Check out #TopoBench, a comprehensive, modular library that benchmarks a wide variety of TDL models across diverse datasets.
Thrilled to see it accepted at DMLR!
Code+Papers:
geometric-intelligence.github.io/TopoBench/
We can’t wait to see what the community builds with TopoBench! Star our repo, read the paper, and bring reproducibility to your TDL research. Contributions and feedback are welcome! #OpenSource #Python #AcademicTwitter
This was a massive team effort!
Developed & led by @levtelyatnikov.bsky.social , @gbg141.bsky.social , Theodore Papamarkou. With major contributions from @Marco Montagna, Mustafa Hajij, Ghada Zamzmi, Michael T. Schaub, @ninamiolane.bsky.social , & @sscardapane.bsky.social , and many others!
