Lenka Tětková

On Dec 2, I’ll present our Nature Communications paper “On convex decision regions in deep network representations” at the #ELLIS UnConference co-located with #EurIPS.
We show that human & machine learning may share the same geometry — convexity.
👉 eurips.cc/ellis
@ellis.eu @euripsconf.bsky.social

24.10.2025 08:16 — 👍 5    🔁 1    💬 0    📌 0

Large Vision Models Can Solve Mental Rotation Problems

Sebastian Ray Mason, Anders Gjølbye, Phillip Chavarria Højbjerg, Lenka Tětková, Lars Kai Hansen

tl;dr: DINOv3,CLIP know 3D geometry, but only in middle layers. MAE bad, ImageNet bad, cls token bad.
arxiv.org/abs/2509.15271

22.09.2025 10:19 — 👍 50    🔁 5    💬 3    📌 2

On convex decision regions in deep network representations - Nature Communications Understanding how machine learning models form internal representations remains a fundamental challenge. Here, the authors explore the geometric property of convexity in latent spaces as a lens to bet...

New work on how human brains and AI systems organize knowledge is out in Nature Communications

www.nature.com/articles/s41...

Great work,
@lenkatetkova.bsky.social and team!

02.07.2025 10:03 — 👍 10    🔁 3    💬 0    📌 0

🧠 New paper alert!
Our study in Nature Communications shows that both human & machine learning may rely on the same geometry: convexity.
It could be a key to shared concept formation.
👉 www.nature.com/articles/s41...
#AI #CognitiveScience #Explainability

02.07.2025 09:10 — 👍 5    🔁 1    💬 0    📌 0

🎥 The recording of the 2nd ELLISxUniReps Speaker Series session with @mtoneva.bsky.social and @lenkatetkova.bsky.social is now available at: youtu.be/qXBT_-w3Mfs

24.06.2025 13:43 — 👍 4    🔁 2    💬 1    📌 1

🎉 Get ready for our second @ellis.eu × UniReps Speaker Series session! 🔥

🗓️ When: June 20, 2025 – 16:00 CET
📍 Where: ethz.zoom.us/j/66426188160
🎙️ Speakers: Keynote by @mtoneva.bsky.social & Flash Talk by @lenkatetkova.bsky.social
👉 Join our Google group: groups.google.com/u/2/g/ellis-...

11.06.2025 11:48 — 👍 7    🔁 3    💬 0    📌 3

Invited Speakers for the XAI4CV workshop at CVPR 2025 — Mihaela van der Schaar, Tsui-Wei (Lily) Weng, Klaus-Robert Müller, Junfeng He, Chinasa T Okolo

We are thrilled to welcome an incredible lineup of invited speakers to the 4th Explainable AI for Computer Vision (XAI4CV) Workshop, held as part of #CVPR2025 — which kicks off next week, from Wednesday, June 11th to Sunday, June 15th in Nashville, TN!

05.06.2025 12:58 — 👍 11    🔁 6    💬 1    📌 0