Erkan Turan

PhD student at @Polytechnique. Prev: @UniofOxford and MVA. Interests: anything interesting

[Erkan Turan @erkanhturan]

Why are GNNs stuck at 2–4 layers? Deep ones “oversmooth”, every node collapses to the same vector.

Our fix needs no residual connections, no normalization, no rewiring. Just swap the activation function.

📍Spotlight #ICML2026 , Seoul

Link to paper at end of thread.

[Erkan Turan @erkanhturan]

@antoine_guedon @nico_dufour @ShuNakamuraW @JiahuiLei1998 @kyoto_vision @akanazawa Congrats well deserved !

[Antoine Guédon @antoine_guedon]

🎉Surflo is a NeurIPS 2026 Oral!
Huge thanks to my coauthors @nico_dufour @ShuNakamuraW @JiahuiLei1998 @kyoto_vision @akanazawa, and to the reviewers and AC.

Code, model and data are released. Check it out if you're into feed-forward 3D and flow matching!
https://t.co/lBcJRgpN2O

[Maks Ovsjanikov @maks_ovsjanikov]

How do we know that we live in Euclidean 3D space?

Henri Poincaré argued that a motionless observer cannot acquire the concept of 3D geometry.

In our paper "SeeSE3: Emergence of 3D Space in Vision Features," we study this via vision foundation models.

https://t.co/XHthzoG90A

[Erkan Turan @erkanhturan]

Started a science substack!

First post: "A 100-Dimensional Orange Is All Peel, and Why Physicists Care." It opens with a piece of fruit, detours through the weirdness of high-dimensional space, and lands on the physics behind steam engines:

https://t.co/wqrRJUW3FK

[Erkan Turan @erkanhturan]

Great first session at #ICML2026 for our Koopman-Flow work. Two more posters today in Hall A: our spotlight work on a bifurcation theoretic lense on oversmoothing in GNNs (Thu 10:30, #2502) and spectral-geometry perspective on latent space alignement (Thu 2:30, #2409). Come by!

[Nicolas DUFOUR @nico_dufour]

We explored the impact of variability sources in generative modeling.

Turns out, we've been neglecting the error bars associated with training variability all along!

We should aim to report results that we are sure of their scientific validity, instead of seed engineering!

[Diego Gomez @DiegoGomezLIX]

🎉 Gaussian Wrapping is accepted to ECCV 2026! Grateful to my incredible collaborators 🙌 See you in Malmö 🇸🇪 #ECCV2026

[Antoine Guédon @antoine_guedon]

What if you could turn any number of photos (3, 8, 15, or even 60) into one clean 3D surface (pts & mesh) with Flow Matching?

Check out our new work, Surflo: Consistent 3D Surface Flow Model with Global State. 🧵
1/n

🔗https://t.co/lBcJRgpfdg

[Erkan Turan @erkanhturan]

Paper: https://t.co/dPctQoGMhy

[Erkan Turan @erkanhturan]

Can a linear operator capture what a flow model does?

Surprisingly, yes, if you lift into the right space. We use Koopman theory to linearize the full generative trajectory of a flow model, unlocking spectral analysis and editing.

📍 #ICML2026, Seoul

Code and paper below.

[Erkan Turan @erkanhturan]

https://t.co/IfUYm9tjhq

Work w/ Ari Siozopoulos, @louis_m01, @jgaubil, @emery_piers and my supervisor @maks_ovsjanikov
.

Come find us at @icmlconf in Seoul! 🇰🇷

[Erkan Turan @erkanhturan]

Bottom line: one-step, parallelizable sampling with competitive FID on CIFAR-10 & FFHQ, plus spectral analysis, inversion, semantic editing, and downstream robustness