Less than two weeks to submit your papers on: 📈 #lowrank adapters and #factorizations 🧊 #tensor networks 🔌 probabilistic #circuits 🎓 #theory of factorizations to the first workshop on connecting them in #AI #ML at AAAI please share! 🔁 👇👇👇 april-tools.github.io/colorai/

If in the ELLIS application portal you select one of these topics: 🎲Bayesian & Probabilistic Learning 🦾Robust & Trustworthy ML 🧠Symbolic Machine Learning 🥽Unsupervised Learning consider listing me as a supervisor and check what we do here: april-tools.github.io

landed on 🦋 it feels fresh! the user handle is nolovedeeplearning

I'll be presenting this paper as a poster on Weds 1600-1730! Come and chat about combining discrete hierarchies with LLMs for opinion summarisation 🥳

My "looking for a postdoc" stickers and I are in Miami for #EMNLP2024! 🤩 Do you have/know of a postdoc for summer/autumn '25 related to interpretability, figlang and/or memorisation (vs generalisation)? Reach out! Looking forward to #GenBench2024 on Saturday and the many many...

🥳 We release educational implementations of recent plug-and-play Sparse Attention methods 🥳 While plug-and-play Sparse Attention offers a powerful way to improve LLM inference efficiency, most implementations rely on complex Triton or custom CUDA kernels to match dense

Another nano.* gem from my amazing student Piotr Nawrot: a repo on SOTA sparse attention methods (both for prefilling and generation) It will also feature in the NeurIPS Conference tutorial "Dynamic Sparsity in Machine Learning" I will give with Andre Martins!

🏆🥳🙌Congratulations to (l-r) Dr Burchell Laurie Burchell, Dr Cardenas and Dr Moghe Nikita Moghe on their recent graduation from the UKRI CDT in NLP School of Informatics, The University of Edinburgh ! All the very best in your respective careers! 🌠 🏆

Thank you for the picture antonio vergari ⚔️ not at #ICLR2025 !!! 🚀🚀🚀

The poster fest I organised was so lively and saw many students showing up to rehearse for the upcoming NeurIPS Conference Thanks #GAIL the #Generative #AI Lab School of Informatics, The University of Edinburgh for sponsoring the event!

Heading to Vancouver soon to attend #NeurIPS2024! Stop by our tutorial and posters 👇

Is sparsity the key to conditional computation, interpretability, long context/generation, and more in foundation models? Find out at my #NeurIPS2024 tutorial on Dynamic Sparsity in Machine Learning with Andre Martins! Followed by a panel with Sara Hooker and Alessandro Sordoni 🧵

Come to our #NeurIPS2024 tutorial on Tuesday Dec 10 9:30 AM to find the answers to these questions (or to ask new questions!) 🔬

📕⬇️ My thesis on 🚫unargmaxable outputs is online! Check it out if you want to learn more about how output layers constrain what neural networks can and cannot predict 👉 era.ed.ac.uk/handle/1842/42…

Next step: excited to be doing research towards AGI (Andreas Grivas's Income) as a postdoc with antonio vergari ⚔️ not at #ICLR2025 and Edoardo Ponti !

I have been elected an ELLIS Scholar! On a related note: it's more urgent than ever for ELLIS to take a leading role in advancing an AI plan for Europe and developing foundation models autonomously. DeepSeek-R1 showed us it's possible. ellis.eu/fellows-schola…

Don't forget to submit to NeSy 2025🐲! We have two deadlines this year, so no excuse to submit (but be careful to be on time with US visa's!)

I have a scholarship for a PhD on efficient memory and tokenization in LLMs at EdinburghNLP! Eligibility: UK home fee status Starting date: from July 2025 onwards. informatics.ed.ac.uk/study-with-us/… Please contact me if you're interested!

Sparse attention is one of the most promising strategies to unlock long-context processing and long generation reasoning in LLMs. We performed the most comprehensive study on training-free sparse attention to date. Here is what we found:

We built sparse-frontier — a clean abstraction that lets you focus on your custom sparse attention implementation while automatically inheriting vLLM’s optimizations and model support. As a PhD student, I've learned that sometimes the bottleneck in research isn't ideas — it's