COLTupf

🔊New EMNLP paper from @elegualdoni & Gemma Boleda! Why do objects have many names? Human lexicons contain different words that speakers can use to refer to the same object, e.g., purple or magenta for the same color. We investigate using tools from efficient coding... 1/3

We are migrating to bluesky and will shortly delete this account! Find us here: bsky.app/profile/colt-u…

Want to use computational tools to figure out how human language works? But not ready for a PhD? UC Irvine's new post-bacc program in computational language science bridges the gap. Fall 2024 applications now open! socsci.uci.edu/newsevents/new…

Louise McNally Honored as an LSA Fellow upf.edu/web/glif/1/-/a…

Next up: Best Extended Abstract Award Oral Presentation! 🎉 @sparse_emcheng presents “Evidence from fMRI Supports a Two-Phase Abstraction Process in Language Models”. 🧠✨

If you're at #NeurIPS2024 today, check out our paper "Bridging semantics and pragmatics in information-theoretic emergent communication" neurips.cc/virtual/2024/p… Poster session: Fri Dec 13, 2-5pm, #3711 w/ @elegualdoni @roger_p_levy @MycalTucker

Presenting our work on "The representation landscape of few-shot learning and fine-tuning in large language models" @NeurIPSConf Poster Session East 1 Wednesday 1 h. 11-14 #3303 Great work with @DiegoDoimo @aleserra1998 @ansuin @AreaSciencePark More below

🚨 PhD position alert! 🚨 I'm hiring a fully funded PhD student to work on mechanistic interpretability at @UvA_Amsterdam. If you're interested in reverse engineering modern deep learning architectures, please apply: vacatures.uva.nl/UvA/job/PhD-Po…

Conclusion: for communication in-context, Lexical systems with a soft mapping between referents and names let speakers maximize communication accuracy while minimizing complexity. Paper: aclanthology.org/2024.emnlp-mai… 3/3

We explored, for a color naming task, why a soft mapping between referents and words is a good solution for communication... ...by taking into account 1⃣in-context communication 2⃣the hierarchical structure of the lexicon 2/3

🔊New EMNLP paper from @elegualdoni & Gemma Boleda! Why do objects have many names? Human lexicons contain different words that speakers can use to refer to the same object, e.g., purple or magenta for the same color. We investigate using tools from efficient coding... 1/3

Find us on bluesky.social ! bsky.app/profile/colt-u…

Our department is looking for a tenure-track faculty member in Language Learning & English Linguistics! apply.interfolio.com/158799 I'll be happy to answer any questions re: my experience at UPF, the department, Barcelona, etc. All help spreading the word will be appreciated😁

✨ Meet #ResiDual, a novel perspective on the alignment of multimodal latent spaces! Think of it as a spectral "panning for gold" along the residual stream. It improves text-image alignment by simply amplifying task-related directions! 🌌🔍 arxiv.org/abs/2411.00246 [1/6]

💡New preprint! Grammar* has many purposes! 1⃣Encode meaning The Spanish word "gatas" is feminine and plural, and means the cats are female, and more than one (⛳️but are all "gatos" male?) 1/6 *noun inflectional/derivational morphology

📢 Seminari del #COLT "Understanding Language Models via Theory, Interpretability, and Humans". 📅 dimecres 23 d'octubre del 2024 🕑 de 12.00 a 13.00 h 🎙️ Michael Hahn (Saarland University) 📍 aula 52.217 del Campus Poblenou - UPF ℹ️ tuit.cat/c90tZ @colt_upf

🚨𝗧𝗮𝗸𝗲𝗮𝘄𝗮𝘆: be careful when quantifying your LM's uncertainty! Use probing-based methods if possible. For more practical takeaways see aclanthology.org/2024.acl-long.… 5/5

⛳️No uncertainty quantification methods were stable to semantics-preserving perturbations! (paraphrase, translation,...) 4/5

Belated update🥳! Matéo Mahaut presented this at @aclmeeting as part of an internship with @lauraina_ @mhardalov at @amazon Paper: aclanthology.org/2024.acl-long.… How can we quantify uncertainty of LLMs? 🚨Not all methods are equally robust... 🧵 1/5