Maybe see you again next year, at Sentosa island, Singapore #emnlp #emnlp 2023 Huiyin Xue Nikos Aletras Sheffield NLP

CMYKCS Halo Bartender | #SaveWarriorNun Bridge Huiyin Xue Henrique 👀 GREY ΔΗΜΗΤΡΑ ΚΑΡΑΚΑΣΗ Arrow Wilson.sdv Yeimy Adle 🇰🇪ג'וזף דה הווארד kiplagat Felix. Trevorphilips.eth Dan E. Hendrickson Juan Vilo Ulloa ラマントス e'ABEL Caroline McIntyre lunare 🍒 GentePlus Narayan Bheel Jodhpur

If you missed Huiyin Xue's #EMNLP2022 poster yesterday on HashFormers, you can find the video/poster/slides here: underline.io/events/342/ses…

Huiyin Xue Huiyin, where is your paper?

HashFormers: Towards Vocabulary-independent Pre-trained Transformers
deepai.org/publication/ha…
by Huiyin Xue et al.
#Computation #Language

Xutan Peng Mark Stevenson Aline Villavicencio Haiping Lu Nikos Aletras Ivan Vulić Sheffield Comp Sci Sheffield NLP Congrats👏!

Danae Sánchez Catalina Goanta Nikos Aletras Congratulations!👏👏👏

Yizhi Li It's UAE Presidential Palace🤖

MHE attention only requires a negligible fraction of additional parameters (3nd, where n is the number of attention heads and d the size of the head embeddings) compared to a single-head attention, while MHA requires (3n^2−3n)d^2−3nd additional parameters.
[4/k]

MHE is substantially more memory efficient compared to alternative attention mechanisms while achieving high predictive performance retention ratio to vanilla MHA on several downstream tasks.
[3/k]

Huiyin Xue Nikos Aletras Many congrats!

Houda BOUAMOR Huiyin Xue Nikos Aletras Sheffield NLP the business meeting this afternoon 😆

Huiyin Xue Thank you Huiyin!🥰

casszhao Huiyin Xue Nikos Aletras Sheffield NLP Where was this announced?

Houda BOUAMOR casszhao Huiyin Xue Nikos Aletras Sheffield NLP Business meeting

Pit One Against Many: Leveraging Attention-head Embeddings for Parameter-efficient Multi-head Attention (findings)

Huiyin Xue, Nikos Aletras