Check out our new preprint 𝐓𝐞𝐧𝐬𝐨𝐫𝐆𝐑𝐚𝐃. We use a robust decomposition of the gradient tensors into low-rank + sparse parts to reduce optimizer memory for Neural Operators by up to 𝟕𝟓%, while matching the performance of Adam, even on turbulent Navier–Stokes (Re 10e5).

I am excited to announce our latest work 🎉 "Cultural Evaluations of Vision-Language Models Have a Lot to Learn from Cultural Theory". We review recent works on culture in VLMs and argue for deeper grounding in cultural theory to enable more inclusive evaluations. Paper 🔗: arxiv.org/pdf/2505.22793

When I grew up in Sweden in the 1990s the government distributed the book “Tell Ye Your Children”. Its purpose is to remind people how fascism, and seeing oneself as better than others can lead to inhuman evil. I think about it often and recommend you read it too.

SCIA 2025 submission deadline extended to Feb. 17

LoQT: Low-Rank Adapters for Quantized Pretraining: neurips.cc/virtual/2024... Authors: @sloeschcke.bsky.social, @madstoftrup.bsky.social, @mkastoryano.bsky.social, @serge.belongie.com, @vesteinns.bsky.social 🗓️ West Ballroom A-D #6104 - Thu 12 Dec 4:30 p.m. PST — 7:30 p.m. PST (2/5)

Happy to share our work "Counterfactual Generation from Language Models" with @AnejSvete, @vesteinns, and Ryan Cotterell! We tackle generating true counterfactual strings from LMs after interventions and introduce a simple algorithm for it. (1/7) arxiv.org/pdf/2411.07180

welcome @ellis.eu! could you please extend the deadline to review #PhD applicants? 🙏 as things are now, you are giving (postdocs, senior PhD students) reviewers 5 working days to evaluate 16 (and more?) applications... otherwise reviewers will do cheap evaluations looking at the usual "metrics"

Let’s welcome @ellis.eu to Bluesky and give them a follow! 🦋

Excited to see LoQT adopted in memory-constrained settings, lowering the barrier to training models such as LLMs! Great collaboration with @mabeto5p, @mjkastoryano, @sergebelongie.bsky.social , @vesteinns.bsky.social Code: github.com/sebulo/LoQT 💻 Paper: arxiv.org/abs/2405.16528 📄