Thanks Jelle!! I'll make sure to prominently feature UvA in the talk tomorrow π all discussions for Compositionality Decomposed were held back at Nikhef!
20.08.2025 03:57 β π 0 π 0 π¬ 0 π 0
Congratulations to the winners of the 2025 IJCAIβJAIR Prize for their paper βCompositionality Decomposed: How Do Neural Networks Generalise?β β Dieuwke Hupkes, Verna Dankers, Mathijs Mul, and Elia Bruni! Presented by Edith Elkind, Northwestern University arxiv.org/abs/1908.08351
#IJCAI2025
19.08.2025 19:11 β π 5 π 1 π¬ 0 π 0
Proud to accept a 5y outstanding paper award
@ijcai.org π from the Journal of AI Research for the impact Compositionality Decomposed has had, on behalf of the team (Dieuwke Hupkes, Elia Bruni & Mathijs Mul)!𧑠Come to room 513 on Wed@11.30 to learn about rethinking compgen evaluation in the LLM eraπ€
19.08.2025 18:27 β π 14 π 1 π¬ 1 π 0
Main takeaway: students excel both by mimicking teachers and deviating from them & be careful with distillation, students may inherit teachersβ pros and cons. Work done during a Microsoft internship. Find me ππΌ Weds at 11 in poster session 4 arxiv.org/pdf/2502.01491 (5/5)
27.07.2025 15:37 β π 0 π 0 π¬ 0 π 0
...identifying scenarios in which students *outperform* teachers through SeqKDβs amplified denoising effects. Lastly, we establish that through AdaptiveSeqKD (briefly finetuning your teacher prior to distillation with HQ data) you can strongly decrease memorization and hallucinations. (4/5)
27.07.2025 15:37 β π 0 π 0 π¬ 1 π 0
Students showed increases in verbatim memorization, large increases in extractive memorization, and also hallucinated much more than baselines! We go beyond average-case performance through additional analyses of how T, B, and S perform on data subgroups... (3/5)
27.07.2025 15:37 β π 0 π 0 π¬ 1 π 0
SeqKD is still widely applied in NMT to obtain strong & small deployable systems, but your student models do not only inherit good things from teachers. In our short paper, we contrast baselines trained on the original corpus to students (trained on teacher-generated targets). (2/5)
27.07.2025 15:37 β π 0 π 0 π¬ 1 π 0
Thrilled to be in Vienna to learn about all the papers & catch up with NLP friends π¦πΉ You'll find me at the ACL mentorship session (Mon 2PM), at our @l2m2workshop.bsky.social (Fri) and at my poster (Wed 11AM)! I'll present work w/ @vyraun.bsky.social on memorization in NMT under SeqKD #ACL2025 (1/5)
27.07.2025 15:37 β π 8 π 0 π¬ 1 π 0
I miss Edinburgh and its wonderful people already!! Thanks to @tallinzen.bsky.social and @edoardo-ponti.bsky.social for inspiring discussions during the viva! I'm now exchanging Arthur's Seat for Mont Royal to join @sivareddyg.bsky.social's wonderful lab @mila-quebec.bsky.social π€©
01.07.2025 21:33 β π 15 π 1 π¬ 3 π 0
A circular diagram with a blue whale icon at the center. The diagram shows 8 interconnected research areas around LLM reasoning represented as colored rectangular boxes arranged in a circular pattern. The areas include: Β§3 Analysis of Reasoning Chains (central cloud), Β§4 Scaling of Thoughts (discussing thought length and performance metrics), Β§5 Long Context Evaluation (focusing on information recall), Β§6 Faithfulness to Context (examining question answering accuracy), Β§7 Safety Evaluation (assessing harmful content generation and jailbreak resistance), Β§8 Language & Culture (exploring moral reasoning and language effects), Β§9 Relation to Human Processing (comparing cognitive processes), Β§10 Visual Reasoning (covering ASCII generation capabilities), and Β§11 Following Token Budget (investigating direct prompting techniques). Arrows connect the sections in a clockwise flow, suggesting an iterative research methodology.
Models like DeepSeek-R1 π mark a fundamental shift in how LLMs approach complex problems. In our preprint on R1 Thoughtology, we study R1βs reasoning chains across a variety of tasks; investigating its capabilities, limitations, and behaviour.
π: mcgill-nlp.github.io/thoughtology/
01.04.2025 20:06 β π 52 π 16 π¬ 1 π 9
π
11.03.2025 01:02 β π 1 π 0 π¬ 0 π 0
Super excited about the First L2M2 (Large Language Model Memorization) workshop at #ACL2025 in Vienna @l2m2workshop.bsky.social! π₯³ Submit via ARR in February, or directly to the workshop in March! Archival and non-archival contributions are welcome. ποΈ sites.google.com/view/memoriz...
27.01.2025 22:03 β π 2 π 0 π¬ 0 π 0
thinking of calling this "The Illusion Illusion"
(more examples below)
01.12.2024 14:33 β π 1585 π 389 π¬ 60 π 91
The largest workshop on analysing and interpreting neural networks for NLP.
BlackboxNLP will be held at EMNLP 2025 in Suzhou, China
blackboxnlp.github.io
PhD-ing at McGill Linguistics + Mila, working under Prof. Siva Reddy. Mostly computational linguistics, with some NLP; habitually disappointed Arsenal fan
Researcher at Google and CIFAR Fellow, working on the intersection of machine learning and neuroscience in MontrΓ©al (academic affiliations: @mcgill.ca and @mila-quebec.bsky.social).
PhD student at the University of Edinburgh.
Co-creator of AlWird (Arabic Wordle).
Research interests: Diversity of Arabic Dialects, Arabic NLP, Multilinguality.
https://amr-keleg.github.io/
Associate Professor at GroNLP ( @gronlp.bsky.social⬠) #NLP | Multilingualism | Interpretability | Language Learning in Humans vs NeuralNets | Mum^2
Head of the InClow research group: https://inclow-lm.github.io/
AI, philosophy, spirituality
Head of interpretability research at EleutherAI, but posts are my own views, not Eleutherβs.
PhD candidate at Tilburg University, doing research on interpretability for text and speech.
https://hmohebbi.github.io/
Visiting Scientist at Schmidt Sciences. Visiting Researcher at Stanford NLP Group
Interested in AI safety and interpretability
Previously: Anthropic, AI2, Google, Meta, UNC Chapel Hill
Postdoc @ CMU Machine Learning and Philosophy β’ Metascience of AI: safeguarding scientific integrity in the age of AI β’ gbalint.me β’ ππΊπ΄σ §σ ’σ ³σ £σ ΄σ Ώ
Postdoc at Uppsala University Computational Linguistics with Joakim Nivre
PhD from LMU Munich, prev. UT Austin, Princeton, @ltiatcmu.bsky.social, Cambridge
computational linguistics, construction grammar, morphosyntax
leonieweissweiler.github.io
https://najoung.kim
langauge
Asst Prof at Johns Hopkins Cognitive Science β’ Director of the Group for Language and Intelligence (GLINT) β¨β’ Interested in all things language, cognition, and AI
jennhu.github.io
I make colorless green GPUs sleep brrriously. Computational phonology, morphology, language change models, speech/language technologies (especially for people with disabilities).
Stanford Professor of Linguistics and, by courtesy, of Computer Science, and member of @stanfordnlp.bsky.social and The Stanford AI Lab. He/Him/His. https://web.stanford.edu/~cgpotts/
Research Scientist at Ai2, PhD in NLP π€ UofA. Ex
GoogleDeepMind, MSFTResearch, MilaQuebec
https://nouhadziri.github.io/
Postdoc @rug.nl with Arianna Bisazza.
Interested in NLP, interpretability, syntax, language acquisition and typology.
NLP, Linguistics, Cognitive Science, AI, ML, etc.
Job currently: Research Scientist (NYC)
Job formerly: NYU Linguistics, MSU Linguistics
