Pinned Tweet
馃毃GNNs are not continous across Graph Resolutions馃毃
#ICML2026
馃У1/5 Should GNNs produce similar representations for the same object at different resolutions?
A social network of individual users vs one of interacting groups.
A molecule with all atoms vs just heavy atoms.
Christian Koke retweeted
Come by poster 706 to discuss discrete diffusion metrics and genppl!
Generative Perplexity, even when normalized by entropy, is not a good metric for discrete diffusion model performance. Come find out why tomorrow afternoon at @antoniofrancaib's SPIGM poster.
Christian Koke retweeted
Drop by our poster on Strong Stochastic Flow Maps later today!
馃搶 11:40-13:30 | Hall A #706
w/@sgmccallum, Timothy Herschell, @Niklas_TR, @AlexanderTong7, @JamesFosterBath
Christian Koke retweeted
Tomorrow, I'm at #ICML2026 presenting our paper at the Structured Probabilistic Inference & Generative Modeling (SPIGM) workshop:
馃搫 "Hacking Generative Perplexity: Why Unconditional Text Evaluation Needs Distributional Metrics"
w/ @AlexanderTong7
馃У馃憞
Tagging some people that might be interested:
@JLutzeyer @PetarV_93 @gasteigerjo @jacobbamb @fedzbar @hxyscott @Pseudomanifold @mmbronstein
Full paper here: arxiv.org/abs/2605.31315







